|
on Big Data |
| By: | Dexin Peng; Xiaoyu Wang |
| Abstract: | Shallow models are special cases of deep models, and deep models theoretically have the potential to outperform the shallow ones. However, the existing empirical asset pricing literature provides strong benchmarks for shallow models. Residual learning allows neural network models in asset pricing to go deeper by preserving and refining their shallow counterparts. The out-of-sample Sharpe ratio for value-weighted long-short portfolios of deep residual models (2.07) is higher than that for the corresponding shallow ones (1.92) and more than twice that of the deep feedforward models (0.89). We show that model depth is a source of additional economic value in asset pricing. Residual learning can be used to deepen other neural-network-based asset pricing models if they contain intermediate layers. Our design also provides one way to scale asset pricing models, making native "large asset pricing models" more feasible. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.09613 |
| By: | Raša Karapandža; Yaw Nyarko |
| Abstract: | Large language models (LLMs) increasingly provide economic and financial advice. We develop a microeconomic framework for understanding this role. We prove that an LLM acts like a voting rule: each input text is an election, possible continuations are candidates, and output probabilities are vote shares derived from the training corpus. Such rules can generate Condorcet cycles, implying intransitive pairwise recommendations and potential money-pump exploitation. Using ChatGPT-4o, we identify millions of cycles in pairwise investment choices among S&P 100 firms. Our theorems establish the voting interpretation on the training data and support its use off-sample. We also show that correcting cycles ex post is generally infeasible. Finally, temperature changes the voting rule: at T = 1, outputs are sampled in proportion to vote shares; lower temperatures favor the plurality winner, whereas higher temperatures flatten probabilities toward uniformity. |
| JEL: | D81 D83 D91 G41 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35839 |
| By: | Thomas Delcey (LEDi - Laboratoire d'Economie de Dijon [Dijon] - UBE - Université Bourgogne Europe); Aurélien Goutsmedt (ICHEC - Brussels Management School [Bruxelles], UCLouvain - Université Catholique de Louvain = Catholic University of Louvain); Alexandre Truc (GREDEG - Groupe de Recherche en Droit, Economie et Gestion - UNS - Université Nice Sophia Antipolis (1965 - 2019) - CNRS - Centre National de la Recherche Scientifique - UniCA - Université Côte d'Azur) |
| Abstract: | This article demonstrates how unsupervised quantitative methods can enrich the history of economic thought. Using the largest English-language corpus ever assembled for the field—nearly 260, 000 economics journal articles from 1900 to 2009 with citation data—we analyze the evolution of the concept of rationality. Combining large language model–based semantic analysis with bibliometric and network methods, we identify and cluster discussions of rationality across time and scales, such as the emergence of a social view of rationality, the circulation of bounded rationality, and the emergence of behavioral economics. We provide an open-source interactive tool to support transparency and reuse. |
| Keywords: | Text mining, large language models, natural language processing, behavioral economics, bibliometrics, economics history, Bounded rationality, rationality |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:hal:journl:hal-05431080 |
| By: | Levi Boxell; Jacob Conway; Nicolas Longuet-Marx |
| Abstract: | What role do journalists play in determining the political slant of the news they produce? We develop and estimate a model where journalists and newspaper outlets contract over both slant and wages. The model implies a set of conditions under which we can consistently estimate the role of journalist preferences in driving the observed variation in slant across outlets by leveraging journalist transitions between outlets. To measure slant, we train a transformer-based, machine learning model using articles tweeted by politicians and apply it to a full-text database of 9+ million newspaper articles published in the US between 2013 and 2018. Under our model and identifying assumptions, our estimates (a) reject the hypothesis that journalists have zero ideological preferences over the content they produce and (b) imply that 10% of the variation in slant can be explained by journalists. When journalists move across outlets, their average slant shifts by 77% of the gap between their destination and origin outlet averages. In counterfactuals that reweight journalist party composition, moving from the observed left-leaning distribution to an even distribution of Democrats, Republicans, and non-partisans shifts average article slant in a more conservative direction, but only by 0.04 standard deviations in slant. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.04182 |
| By: | OECD |
| Abstract: | Timely and accurate distributional income data are essential for monitoring inequality, assessing the adequacy of tax-benefit systems, and designing policy responses during periods of rapid economic change. However, income and inequality statistics derived from EU-SILC are typically published with a lag of 12 to 18 months, limiting their usefulness during fast-moving shocks such as the COVID-19 pandemic, the 2022 energy price shock, or a severe recession. This paper presents results from a new nowcasting framework designed to reduce this information gap for EU Member States. By combining machine-learning (ML) forecasts of market income with EUROMOD microsimulation under the latest available tax-benefit rules, the framework produces timely estimates and forecasts of income inequality. |
| Keywords: | inequality, machine learning, microsimulation, nowcasting, poverty rate |
| JEL: | D31 C53 |
| Date: | 2026–10–13 |
| URL: | https://d.repec.org/n?u=RePEc:oec:wiseaa:41-en |
| By: | Isha Agarwal; Wentong Chen; Eswar S. Prasad |
| Abstract: | When does bad news stick? This paper shows that the persistence of media sentiment depends not only on whether news is good or bad, but also on what the news is about. Using the complete Wall Street Journal archive from 1990 to 2025, we construct monthly positive and negative sentiment indices for twelve economic topics and estimate topic-specific sentiment dynamics. Bad-news innovations generate larger cumulative responses than positive-sentiment innovations in every topic, indicating that negativity asymmetry is a broad feature of media sentiment. But the size of this asymmetry varies sharply across topics. The bad-minus-positive persistence gap is largest in security and human rights, firms and corporate strategy, health and public safety, and real estate and financial stability. Bad-news persistence is also sizable in data-rich topics such as financial markets, real estate, and corporate news, suggesting that frequent information arrival does not necessarily lead narratives to revert quickly. These findings show that aggregate sentiment indices mask substantial topic-level heterogeneity in the asymmetric dynamics of narratives, with potential implications for how persistent negative sentiment shapes expectations, asset prices, and real economic outcomes. |
| JEL: | C55 D83 D84 G14 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35838 |
| By: | Mateusz Buczy\'nski; Micha{\l} Wo\'zniak; Konrad Kaczy\'nski; Anna Wr\'oblewska; Sebastian Kuk |
| Abstract: | Forecasting resale prices of used electronics is critical for subscription-based platforms where pricing errors translate directly into risk. Unlike structured financial markets, second-hand electronics exhibit high volatility, sparse listing histories, and non-normal price dynamics - yet no systematic time-series benchmark exists for this domain. This paper presents the first multi-horizon benchmark of statistical and deep learning forecasting models for used electronics price prediction. We use a large-scale dataset of daily price listings from Polish online marketplaces (January 2022 to March 2025, 100+ smartphone and laptop models) and evaluate eleven models across six horizons from 1 to 365 days, covering classical methods (ARIMA, ETS, Theta), recurrent and convolutional networks (LSTM, TCN), and modern deep architectures (N-BEATS, N-HiTS, TFT, PatchTST, Informer). Three complementary evaluation protocols assess trajectory fitness, one-shot endpoint accuracy, and cross-horizon transfer. N-BEATS achieves the lowest MAPE beyond 30 days, reaching 8.51% at 365 days versus 14.94% for the best statistical baseline - a 43% reduction. At short horizons (1-7 days), all models converge near 0.72% MAPE and the naive baseline remains competitive. A single N-BEATS model trained at 365 days generalizes to all shorter horizons, eliminating the need for horizon-specific models. N-BEATS and N-HiTS also demonstrate superior hyperparameter stability. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.10727 |
| By: | Graeber, Daniel (DIW Berlin); Meister, Lorenz (The Institute for New Economic Thinking at the Oxford Martin School (INET Oxford), University of Oxford); Murray, Neil (DIW Berlin) |
| Abstract: | Terrorism affects individuals far beyond its immediate victims, with large economic consequences. We estimate the effect of terrorism on risk preferences by merging geocoded panel data with the Global Terrorism Database in a staggered difference-in-differences design comparing those within and beyond 25 kilometers of attacks in Germany. Exposure reduces willingness to take risks by 2 percent of a standard deviation—localized, peaking after one year, dissipating within five—implying welfare losses near 106 million Euro annually. We classify 60, 000 news articles with a fine-tuned language model and separate two media channels: the salience and the emotional tone of coverage. Holding attack severity fixed, responses are concentrated among attacks with more extensive and more negative coverage. The survey data corroborate media transmission—respondents who do not read newspapers do not respond—and point to declining happiness as the emotional mechanism. After severe attacks, self employment falls by 2 percentage points. |
| Keywords: | Terrorist attacks, risk preferences, staggered difference-in-differences |
| JEL: | D12 D81 D91 I31 |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:amz:wpaper:2026-27 |
| By: | Josh Lerner; Fiona Paine; Antoinette Schoar |
| Abstract: | Private capital fund managers appear to possess significant information about the relative quality of their portfolio firms even within the first three years after the investment. But the way this information is disclosed to investors changes dynamically: Textual discussions of portfolio companies have substantial information about the firms’ ultimate performance in the early stages of investment, while interim valuations are less important. Over time, this pattern reverses, with interim valuations becoming more important in explaining future returns, especially after portfolio companies have a financing event. The results vary not only with the tone but also with the content of fund manager disclosures. In an opaque market with substantial information asymmetries, textual discussions seem to serve as a signal about general partner quality. |
| JEL: | G24 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35860 |