|
on Computational Economics |
| By: | Hrishikesh Singh; Aarti Sharma; Millie Pant |
| Abstract: | In modern financial markets, more than 3/4th of the large volume trade is executed by algorithmic trading based on advanced trading strategies to attain higher precision and speed than traditional human traders. The conventional strategies focused on price-based indicators are inadequate in capturing the market dynamism arising from real-world events such as social sentiments, political instability, or sudden economic shifts. With the rise of Deep Reinforcement Learning (DRL) and data availability, there is a requirement for more efficient algorithms that can incorporate these external factors in trading strategies. DRL has shown potential in simulating real-world scenarios and in simplifying complex decision-making since DRL agents can automate the trading process i.e. executing trades in a manner comparable to human traders. They have shown reliable performance since they can update their policies based on risk and quantitative factors. In this paper, we propose a DRL approach based on partial and deep-infusion of sentimental analysis. It features distinct policy designs of actor-critic-based agents: only policy optimization (A2C & PPO) and policy optimization along with Q-learning principles (DDPG, TD3 & SAC). Our findings reveal, that the agents based on policy optimization combined with Q-learning, significantly and consistently demonstrate improved performance upon sentiment deep-infusion in profit maximization as well as risk-averse trading. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.32237 |
| By: | Reichel, Felix |
| Abstract: | Prompt engineering has developed as a transformative subfield in artificial intelligence (AI), enabling humans to enhance AI systems through structured (PROLOG-like) instructions to perform specific tasks. This paper shines a light on its multi-dimensional impact on programming, routine tasks and broader AI applications, drawing on insights from recent studies. By using techniques such as parameter-efficient fine-tuning (PEFT), zero-shot learning, multi-shot learning and linguistic feature analysis, prompt engineering often is used to tackle remaining issues in software engineering (SE), language processing (NLP) and AI safety. The body of literature highlights its role in optimizing AI performance, reducing computational overhead and reducing biases. Additionally, it highlights critical gaps in current methodologies, such as the comparative energy intensity of state-of-art large language models (LLMs) and the requirements for adaptive and scalable prompt crafting strategies. Finally, the paper suggests future directions for research, highlighting the importance of human-centered design and collaboration to advance the field. |
| Date: | 2026–10–07 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:tqf95_v1 |
| By: | Giorgos Iacovides; Oliver Hannibal; Wuyang Zhou; Danilo Mandic |
| Abstract: | Recent advances in Generative AI have substantially improved financial sentiment analysis through post-trained financial large language models (LLMs). However, existing approaches remain confined to predicting the sentiment of each article in isolation and operate as black boxes, which limits contextual understanding and hinders deployment in high-stakes financial decision-making. To this end, we introduce FinLLM-KG, a model-agnostic framework that combines causal LLMs with a financial knowledge graph, in order to inject temporally grounded, company-level knowledge from SEC 10-K filings at inference time, without updating model weights. Within FinLLM-KG, context is supplied either as a natural-language company profile or as a small set of relevance-ranked factual triples. This enables sentiment predictions to account for broader firm-level information rather than the article alone, at almost zero computational cost and for any causal LLM. In addition, attributing each article-level sentiment score to individual sentences and to the retrieved knowledge-graph facts makes the evidence behind the prediction inspectable, thereby enhancing explainability and supporting human-in-the-loop review. Simulations demonstrate that FinLLM-KG consistently improves portfolio performance over the state-of-the-art baseline. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.36083 |
| By: | Jin Huang; Diego Ferreras Garrucho; Yutong Xie; Walter M. Yuan; Qiaozhu Mei; Chen Lian; Jonathon Hazell |
| Abstract: | Large language models (LLMs) have the potential to meet a key goal in economics: a quantitative model of household decision making, across a variety of settings. Yet existing evaluations cover few surveys and outcomes, and do not study how households adjust to changing economic conditions. We introduce a new evaluation, HouseholdBench, which unites 6 U.S. household surveys and 32 prediction tasks spanning numeric, categorical and probabilistic outcomes, related to consumption, income, labor, expectations, and housing. Using past behavior, demographics and macroeconomic conditions, the tasks test whether LLMs predict behavior, including how households adjust to changes in various policies. We evaluate 13 proprietary and open-weight LLMs against a no-change baseline and a gradient boosted tree model. Most LLMs outperform the no-change baseline, including for policy response tasks—with the best model lowering error for numeric outcomes by 12.2%. Across most tasks, gradient-boosted trees rank first; leading proprietary LLMs approach their performance, but open-weight models lag. LLMs exhibit systematic over- and underprediction across different tasks. We identify methods that enable a 4 billion parameter open-weight model to match proprietary models’ performance: fine-tuning and aggregating 16 predictions per observation. Improvements generalize to policy-response tasks, which are excluded from fine-tuning. We release our datasets, code, and leaderboard on our website. |
| JEL: | C45 C53 D12 E21 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35927 |
| By: | Daiya Mita (Nomura Asset Management Co, ltd.,); Akihiko Takahashi (School of Interdisciplinary Mathematical Sciences, Meiji University /Center for Advanced Research in Finance (CARF), The University of Tokyo) |
| Abstract: | This study proposes a novel equity investment strategy that effectively integrates artificial intelligence techniques, multi factor models and financial technical indicators. In investment-related finance research, there exists a serious problem that many actively managed funds fail to outperform simple index buy-and-hold strategies. Our objective is to address this issue by designing a strategy that not only delivers high investment performance but also maintains interpretability, an often-overlooked aspect in complex models based on artificial intelligence. Specifically, as an equity long (buying) strategy, this paper extends a Five Factor model, a well-known finance model for its theoretical grounding in financial economics and its high accountability, to predict future returns by using a gradient boosting machine and a state space model. In addition, an index futures short (selling) strategy for downside hedging is developed with IF-THEN rules and three technical indicators. To the best of our knowledge, this is the first study to develop an equity investment framework based on a new, more predictable Five Factor model, enhanced by artificial intelligence techniques and technical indicators. Finally, empirical analysis is conducted using the data on the Japanese stock market from 1/04/2010 to 12/29/2023 and shows that the proposed strategy outperforms not only the baseline buy-and-hold strategy, but also typical mutual funds for the Japanese equities. Furthermore, statistical tests for Sharpe and Sortino ratios reveal that while the baseline strategy fails to reject the null hypothesis, particularly in the most recent period, our proposed strategy exhibits statistically significant performance across all subperiods, underscoring its robustness. |
| Date: | 2026–10–23 |
| URL: | https://d.repec.org/n?u=RePEc:cfi:fseres:cf632 |
| By: | Adrian Gadient; Delia Haag; Nicolas Gygax |
| Abstract: | Housing costs are an important indicator of affordability and financial pressure on households. This project develops a housing cost estimation model for the canton of Aargau by combining survey data (Structural Survey 2019–2022) with public register data (e.g., population and buildings). To capture the relationship between predictor variables and monthly net rent we used random forest and an elastic net regression models. The random forest regression model achieved the better out-of-sample performance (RMSE = CHF 315, MAE = CHF 224, R2 = 0.55). The most important predictor was floor area, followed by building construction period and number of rooms. Model performance was stable across regions and dwelling sizes. The geographical variation of the monthly net rent corresponded to patterns known from surveys and market analyses. Robustness checks indicated that the results were stable across alternative model settings. These findings suggest that data from public registers and surveys can be effectively combined to estimate housing costs at scale. Since this approach relies on existing data, it avoids costly additional data collection and is therefore useful for public administration, research, monitoring, and planning. Since the necessary data are available for all Swiss cantons, the model can be applied nationwide. |
| Keywords: | housing costs, rent estimation, machine learning, household budget, affordability |
| JEL: | R21 R31 C52 C53 |
| Date: | 2026–10–13 |
| URL: | https://d.repec.org/n?u=RePEc:bss:wpaper:63 |
| By: | Severin Joho; Enzo Rossi |
| Abstract: | We compiled and processed a novel, comprehensive dataset of 1, 063 speeches delivered by high-ranking Swiss National Bank officials between 1928 and 2024. Beyond tracking their quantitative evolution, we conducted an in-depth analysis of the length, readability and thematic content. To ensure robust results, we benchmarked traditional Python-based NLP text mining against generative artificial intelligence using ChatGPT. Our findings highlight four primary policy insights. First, the frequency of annual speeches has followed an upward trajectory since the mid-1950s, accelerating notably after the transition to flexible exchange rates. Second, a clear shift toward brevity is evident. Third, the language has become increasingly accessible since the 1960s. Fourth, inflation has moved to the forefront of the thematic discourse in the era of flexible exchange rates, while cyclical economic development and unemployment have received less emphasis. Finally, we outline key methodological lessons for deploying LLMs to analyze central bank communication, emphasizing the complementary relationship between probabilistic language models and deterministic Python frameworks, while addressing inherent algorithmic pitfalls and identifying critical open questions for future research. |
| Keywords: | Central bank speeches, Linguistic Analysis, Natural Language Processing (NLP), ChatGPT |
| JEL: | E52 E58 D83 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:snb:snbwpa:2026-13 |
| By: | Matin Tavakoli; Seyedreza Tavakoli; Faezeh Salmani; Amir Amini; Mohammad Hosein Rohban; Gholamali Aminian; Ali Habibnia |
| Abstract: | Financial reinforcement learning often translates an investment objective into an additive risk-adjusted reward. An agent can optimise that reward accurately yet implement a different financial preference. We develop a decision audit for terminal mean-variance equilibrium that separates reward misspecification from learning error. In a finite exogenous-state market, we bound the gain from changing one action using actor, reward and continuation-critic residuals. The certificate ignores state-constant critic offsets, accommodates convex action constraints, and has an independent-sample extension that does not require the equilibrium policy. A continuation-aware actor-critic method estimates the future-policy correction from samples. Across nine controlled markets and twenty seeds, it reduces policy distance by and one-step deviation gain by relative to a matched myopic-reward network. A separate PPO comparison tests the same reward intervention within a standard optimiser. Historical tests cross seven covariance estimators and 10, 30 and 49 US industry portfolios over 432 months: joint reward calibration preserves each fitted mean-variance target, but does not improve realised investment performance. Together, these results provide a way to quantify whether a learned decision implements its stated financial objective, separately from how well that objective performs in a backtest. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.34153 |
| By: | Tiantian Wu; Peng Yang |
| Abstract: | Quantitative investment requires alpha factors that are both predictive and diverse. However, existing factor generation methods, including formulaic mining and neural end-to-end models, are mostly based on a single-agent design, which limits exploration of the factor space and reduces adaptability to changing markets. Multi-agent approaches can broaden exploration, but agents trained under a shared objective often collapse to similar outputs, while existing diversity techniques either scale poorly with asset-universe size or address redundancy only after training. We propose a multi-agent sequential optimization (Det-MASO) framework for generating diversified, context-adaptive alpha factors. It introduces a differentiable log-determinant penalty on the inter-agent correlation matrix directly into the joint objective, discouraging correlated agent behavior during training with computational cost that scales linearly with the number of assets. We further show that this penalty prevents homogeneous equilibria and preserves factor diversity throughout optimization, while a context-conditioned scheduler dynamically adjusts agent emphasis across market regimes and Shapley interaction index analysis is used to characterize the cooperation structure that emerges among them. Experiments show that Det-MASO outperforms baselines, improving the information coefficient by 1.1\%, the annualized return by 6.4\%, and the Sharpe ratio by 0.33. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.34936 |
| By: | Adi Tarabeih |
| Abstract: | Prospect theory splits risky choice into three parts: loss aversion, measured by the relative weight of losses to gains, $\lambda$ (e.g., $\lambda = 2$ implies that losses weigh twice as much as gains), diminishing sensitivity to the size of outcomes (i.e., the curvature and linearity of the value function), and the nonlinear weighting of probabilities. I examined whether Large Language Models (LLMs) display the same properties as people by meta-analyzing 102 fitted estimates from 11 studies covering 32 language models. The three components emerge differently. Loss aversion is modest: the pooled $\lambda$ is 1.26 (95% CI 1.06 to 1.46), far below the strongest human meta-analytic estimate ($\lambda = 1.96$) and compatible with more conservative ones (1.07 and 1.31). Matched by design, LLMs are compatible with people when gains and losses are symmetric ($\lambda = 1.15$ against 1.07), but well below them when losses are smaller than gains (1.46 against 2.3), where human loss aversion is strongest. Value functions are close to linear rather than concave under the Tversky-Kahneman method (1.01, against a human value of 0.67), but more concave than in people under the Tanaka method (0.52, against 0.67), so diminishing sensitivity depends on how it is measured. Probability weighting shows a human-like inverse S shape ($\gamma = 0.81$ under Tversky-Kahneman and 0.84 under the Tanaka method), though on few studies. In sum, LLMs show little loss aversion and little diminishing sensitivity, but retain human-like probability distortion, yet these generalizations shift when looking at specific model families and fitting methods. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.33103 |
| By: | Scielzo Ortiz, Fabio; Grané Chávez, Aurea; Díaz Gorfinkiel, Magdalena |
| Abstract: | Measuring digital polarization around geopolitical conflicts raises two methodological problems that are rarely addressed jointly: how to extract sociologically meaningful constructs from unstructured text at scale, and which representation of that text allows unsupervised methods to recover ideological structure. We study both on a corpus of 76, 816 English-language Reddit comments on the Gaza conflict, collected from 18 subreddits over one year. Five discursive variables (political stance, sentiment, discourse tone, thematic frame and argument quality) are extracted with a large language model through zero-shot, rule-based prompting and validated against expert-coded ground truth. We then benchmark five clustering configurations that cross the feature space (validated discursive variables, dense semantic embeddings, or both) with the clustering architecture (standard k-means versus a robust mixed-data framework combining a Robust Generalized Gower distance with Fast k-medoids), and compare them through a Pairwise Separability Index that measures how distinguishable every pair of clusters is on the discursive variables. Embedding-based configurations conflate rhetorical register with political stance and fail to recover the ideological axis, a failure that persists when 90% rather than 50% of embedding variance is retained. Restricting the feature space to the validated variables is necessary but not sufficient: on the same variables, k-means yields an asymmetric partition that leaves one ideological bloc internally undifferentiated, whereas the robust mixed-data framework recovers a symmetric four-group structure. Substantively, the resulting partition reveals a bidimensional structure in which both the pro-Israel and the pro-Palestine blocs split into a non-hostile faction and a hostile, reactive one, so that ideological position and deliberative-affective register operate as distinct structuring dimensions of the debate. |
| Keywords: | Computational Social Science; Llm Feature Extraction; Mixed-Data Robust Clustering; Gaza Conflict; Polarization; Public Opinion Analysis |
| Date: | 2026–10–14 |
| URL: | https://d.repec.org/n?u=RePEc:cte:wsrepe:51001 |
| By: | Mahony, Christopher Brian; Vemuru, Varalakshmi; Rahim, Aly; Owen, Daniel P. |
| Abstract: | AI- and machine learning–based social risk forecasting and explanation are applied across fragile and climate-affected contexts to predict change in conflict (DRC), population movement (Horn of Africa), and crime (a SIDS), using large multimodal datasets (satellite imagery, NLP on news/social media, economic and climate indicators, geospatial data). Three proof-of-concept models achieve out-of-sample accuracies of 63–76% (conflict) and 70–74% (population change), and generate crime risk proxies where official statistics are scarce. Key associated factors include language about sensitive topics and political actors, economic pressures and prices, climate stress, and social perceptions. Results illustrate how social science–informed AI complements conventional analytics and supports policy and operations for risk monitoring, early action, and targeted resource allocation |
| Date: | 2026–04–01 |
| URL: | https://d.repec.org/n?u=RePEc:wbk:hdnspu:209567 |