|
on Big Data |
| By: | Borgioli, Stefano; Gallo, Giampiero M.; Ongari, Chiara; Mavillonio, Maria Saveria; Giannetti, Caterina |
| Abstract: | Measuring sentiment from financial news is a central task in economics and finance, yet most existing indicators rely on dictionary-based approaches that infer sentiment from word counts and only partially capture context, negation, and semantic structure. This paper proposes a framework for constructing daily news mood indices using transformer-based language models and evaluates whether they better represent sentiment than dictionary-based alternatives. Using 143, 755 financial news articles from Factiva, we classify sentiment at the sentence level with FinBERT and aggregate these predictions into article-level and daily sentiment measures through alternative normalization schemes. We compare the resulting indices with benchmark measures based on Shapiro et al., 2022 and Barbaglia et al., 2025. A central contribution is the validation of alternative sentiment measures against human judgments. We conducted an incentivized annotation exercise in which 444 participants evaluated a validation subsample of 588 financial news articles. Consensus ratings from independent human evaluations serve as an external benchmark for assessing the quality of automated sentiment measures. Across correlation, regression, and classification exercises, transformer-based measures show stronger agreement with human judgments than vocabulary-based alternatives and perform substantially better in distinguishing positive, neutral, and negative articles. Overall, the results suggest that incorporating contextual information through transformer-based language models produces sentiment measures that more closely reflect human assessments of financial news. JEL Classification: C55, C81, E32, E37, G14 |
| Keywords: | daily mood indices, financial news sentiment, FinBERT, human validation, text-as-data, transformer-based language models |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbwps:20263283 |
| By: | Louis-Alexandre Bayol (Banque de France) |
| Abstract: | Using Google Search Volumes (GSV), I develop the GSV-based Digitization of Payments Index (GSVDPI), a multidimensional metric built via machine and deep learning models, to track the evolution of digital payments. This index allows for quantifying the changes in the use of digital payment instruments (card, contactless, mobile, and others) in France and comparison with other European countries like Germany, Italy, and Spain. Compared with standard payment digitization statistics published by institutions like the European Central Bank (ECB), the GSVDPI offers distinctly valuable and real-time insights. Consequently, the GSVDPI accounts for the complex dynamics and policy expectations of the digitization process, which are typically overlooked by univariate metrics. I contribute to the literature by identifying the economic mechanisms driving digital payment adoption and examining its macroeconomic effects through a cross-country analysis. I show that incorporating the GSVDPI into the explanatory variables of a multivariate cash demand framework significantly enhances model performance, thereby offering deeper insights into future payment trends. |
| URL: | https://d.repec.org/n?u=RePEc:boc:carp26:11 |
| By: | LiYang Wang (Washington University in St. Louis); Zhen Zhong (Georgetown University); Zhen Tian (University of Glasgow); Keyu Chen (Wuyi University); Keyu Chen (Wuyi University) |
| Abstract: | Advancements in data fusion and real-time analytics technologies have opened new avenues for addressing complex domain challenges. Financial risk early warning systems often suffer from inefficiency due to information silos and monitoring delays. This paper proposes a credit risk early warning system based on heterogeneous information fusion. The system employs a model architecture integrating deep neural networks and attention mechanisms to extract multidimensional features from diverse data sources such as transaction behaviors and social networks, thereby establishing an early identification mechanism for corporate and individual credit risks. System testing demonstrates that this approach significantly enhances the accuracy and timeliness of risk warnings, outperforming traditional rule-based engine solutions. The findings offer innovative insights for early intervention in financial risks, holding practical significance for safeguarding financial stability. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.15744 |
| By: | Moore, Alexander K.; Li, Bingqing; Wang, Shane; Thomson, Matt |
| Abstract: | Large language models (LLMs) are increasingly adopted in the social sciences as surrogates for human coders. Standard practice validates them by their correspondence with human measurements of the same stimuli using statistics like correlation. However, overall correspondence can conceal biases that distort specific claims. We demonstrate this with open-weight LLM ratings of 568 food and 1, 811 scene photographs. Ratings inflate, attenuate, or reverse effects, and improving overall correspondence with prompts can improve one contrast while worsening another. Rankings also elevate some categories and demote others beyond what random errors of the same size typically produce. Across six prompts, some of these errors are shared, so sampling across prompts may continue to mask biases. We propose validating LLM coders against specific claims they will test rather than the general coding task. |
| Date: | 2026–09–21 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:dz3qe_v1 |
| By: | Miana Plesca (University of Guelph) |
| Abstract: | I compare double/debiased machine learning (DDML) for estimating conditional average treatment effects (CATEs) in Stata and Python. The focus is practical: what Stata users can do directly with the community-contributed ddml package, official Stata CATE tools, and related commands; where Python or R can add value; and how these environments can be combined in applied work. I illustrate how partially linear and interactive DDML models can be used to estimate heterogeneous treatment effects, reducing regularization and overfitting bias through orthogonalization and cross-fitting. I then compare alternative approaches to estimating the treatment-effect function itself, including structured heterogeneous PLM specifications, R-learners, DR-learners, and causal forests. Empirical illustrations use random-assignment data from the Job Training Partnership Act (JTPA) and the National Supported Work (NSW) demonstration, as well as census-based wage applications. I emphasize that machine learning can improve the estimation of control functions and treatment effects, but this does not replace credible identification or careful choices about learners, tuning parameters, and interpretability. I argue that Stata is currently especially useful for transparent DDML estimation and reporting of heterogeneous treatment effects, while Python and R can be useful complements for implementing a broader set of more flexible DDML-CATE meta-learners. |
| URL: | https://d.repec.org/n?u=RePEc:boc:carp26:08 |
| By: | Rishabh Choudhary; Chetan Dave; Chetan Ghate |
| Abstract: | Managing rapid economic growth in India naturally brings into focus the role of inflation forecasts for monetary and fiscal policy making. We investigate whether any of a large set of forecasting models improves upon a univariate auto-regression in forecasting core consumer price inflation. Using a monthly panel of forty macroeconomic and financial indicators we estimate nine classes of models that range from an auto-regression to various factor models, quantile regressions and machine learning specifications. In doing so, we also account for inflation expectations and climate change variables. Our root mean square forecast error model comparison metric operates at horizons of one, three, six and twelve months. With respect to the conditional mean, no model produces a forecast error significantly below that of an auto-regression at any horizon over the full comparison sample. With respect to the conditional distribution, quantile regressions that include estimated factors reduce forecast loss relative to a specification without such factors at every quantile and horizon. Our approach is general enough to be applicable to forecasting core inflation in other emerging market economies. |
| Keywords: | inflation forecasting in EMEs, core inflation, diffusion index models, quantile regression, machine learning, inflation targeting |
| JEL: | C22 C53 E31 E37 O23 |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:een:camaaa:2026-83 |
| By: | Anna Kurowska (University of Warsaw, Faculty of Economic Sciences, Interdisciplinary Centre for Labour Market and Family Dynamics (LabFam)); Magdalena Grabowska (University of Warsaw, Faculty of Economic Sciences, Interdisciplinary Centre for Labour Market and Family Dynamics (LabFam)); Maciej Świtała (University of Warsaw, Faculty of Economic Sciences); Beata Osiewalska (University of Warsaw, Faculty of Economic Sciences, Interdisciplinary Centre for Labour Market and Family Dynamics (LabFam); Cracow University of Economics) |
| Abstract: | While studies highlight COVID-19's disruptive impact on fertility, whether pandemic-era childbearing predictors remain informative post-crisis is less clear. Using Polish Familydemic panel data from 2021–2024 (N = 1, 925, aged 18–49), we apply theory-guided machine learning to evaluate a broad set of individual, employment, partnership, and family characteristics. Because short-term fertility intentions prove most predictive of childbearing, we model both actual births and firm positive intentions using random forest. The pandemic-trained model retained strong post-pandemic predictive performance (AUC-ROC = 0.86), demonstrating that pandemic-era drivers remain informative. Childbearing and intentions shared 16 of their top 20 predictors, highlighting common relational, demographic, and work–family factors. Relationship satisfaction, work–life balance, and domestic work division were particularly prominent, whereas prolonged school and childcare closures lowered predicted probabilities for both outcomes among parents. Ultimately, post-pandemic fertility dynamics reflect persistent constraints embedded in everyday family and working lives. |
| Keywords: | childbearing, fertility intentions, COVID-19 pandemic, machine learning, relationship satisfaction, childcare closures |
| JEL: | J13 J12 J22 C53 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:war:wpaper:2026-32 |
| By: | Fonseca, Luís; Martorana, Giulia; Schupp, Fabian; Trebbi, Giovanni |
| Abstract: | Theory suggests inflation risk premia are positive when supply shocks are expected to dominate demand shocks and negative otherwise. We measure these beliefs using demand and supply narratives derived from inflation news via Causality Extraction, which identifies causal relations between inflation and its drivers. Using narrative extracted from inflation news from the Financial Times for the Euro Area and the Wall Street Journal for the US, our key variable, NetDemand, measures the difference in articles attributing inflation to demand versus supply factors. Consistent with asset pricing theory, inflation risk premia are inversely related to NetDemand across maturities in both regions. This relationship holds even after controlling for the composite PMI and VIX, strengthens with risk aversion in the US and inflation volatility in the Euro Area, and is not subsumed by other measures of demand and supply contributions to inflation, views of professional forecasters, or narratives obtained from LLMs. JEL Classification: C53, E44, G12 |
| Keywords: | demand and supply, natural language processing, news media |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbwps:20263288 |