|
on Computational Economics |
| By: | Zonghao Yang |
| Abstract: | Counterfactual analysis aims to predict potential outcomes under hypothetical scenarios, offering valuable insights for decision-making. This paper investigates the application of large language models (LLMs), specifically the GPT-3.5 model, for counterfactual analysis. We focus on the online lending context, where the counterfactual return on investment (ROI) is crucial for evaluating different interest rate schemes. We begin by assessing the predictive performance of GPT and comparing it with advanced machine learning algorithms. The results show that prompt engineering can significantly enhance GPT's predictions, with the R-squared increasing from 1.97% to 2.84%, closely approaching the 3.48% achieved by gradient-boosted regression. Subsequently, we utilize GPT to generate counterfactual ROIs under a set of alternative interest rates. GPT exhibits logical coherence and causal reasoning in its responses. The findings underscore the potential of LLMs as effective tools for counterfactual analysis in online lending, suggesting broader applications for LLMs in various predictive and decision-making contexts. |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2608.05367 |
| By: | Mingli Chen (University of Oxford); Rama Cont (University of Warwick); Andreas Joseph (Bank of England); Michael Kumhof (Bank of England); Xinlei Pan (University of California (Berkeley)); Wei Xiong (University of Oxford); Xuan Zhou (Reserve Bank of Australia) |
| Abstract: | We propose deep reinforcement learning (DRL) as a general approach to bounded rationality in dynamic stochastic general equilibrium (DSGE) models. Agents are represented by deep artificial neural networks and learn to maximise their intertemporal objective function by interacting with an a priori unknown environment. Applying this approach to a model from the adaptive learning literature, DRL agents can learn all equilibria irrespective of local stability properties. However, learning is slow and may be unstable without the imposition of early stopping criteria. These findings can have implications for the use and interpretation of DRL agents and of DSGE models more generally. |
| Keywords: | Artificial intelligence;deep reinforcement learning;adaptive learning;monetary policy;fiscal policy;multiple equilibri |
| JEL: | C14 C52 D83 E52 E62 |
| Date: | 2025–09–26 |
| URL: | https://d.repec.org/n?u=RePEc:boe:boeewp:023264 |
| By: | Giovanni Dispoto; Marcello Restelli; Carmine Ventre |
| Abstract: | Modern portfolio management increasingly demands a balance between traditional risk-adjusted returns and strict Environmental, Social, and Governance (ESG) mandates. Current Reinforcement Learning (RL) approaches typically optimize for a single ESG provider, neglecting the significant divergence in rating methodologies across the industry and the unintuitive nature of manually weighting conflicting objectives. This paper addresses these limitations by formulating ESG-aware portfolio optimization as a Multi-Objective Reinforcement Learning (MORL) problem that simultaneously incorporates ratings from three distinct ESG agencies. To bridge the gap between high-dimensional algorithmic trade-offs and human decision-making, we integrate a Preference Elicitation framework using Gaussian Processes. This system enables practitioners to infer their latent utility functions through intuitive pairwise comparisons of candidate portfolios based on their Sharpe ratios and aggregate ESG scores. We systematically evaluate our framework by employing Large Language Model (LLM) personas to simulate Portfolio Managers operating under varied regional contexts. Empirical results using historical market data reveal that regional backgrounds fundamentally shift the derived preference weights. For instance, European-based personas tend to prioritize ESG alignment over financial returns, while Texas-based personas favor risk-adjusted performance. This work offers a highly adaptable framework that successfully aligns multi-objective algorithmic trading with diverse, real-world human sustainability preferences. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.02677 |
| By: | Harris Cobb; Wenbo Hao; Yingjie Liu |
| Abstract: | We present a novel application of Neural Networks with Local Converging Inputs (NNLCI) to improve the efficiency of existing numerical methods for pricing multi-asset options. The most concise input format for NNLCI has been introduced, offering substantial convenience and efficiency. NNLCI uses a neural network to locally correct solutions from a coarse mesh and a refined mesh (relative to the coarse one), requiring only a minimal amount of high-fidelity training data. We demonstrate this approach on cash-or-nothing options under the Black-Scholes equation in one, two, and three spatial dimensions, and on single-asset down-and-out barrier call options under the Heston stochastic-volatility model (whose pricing PDE is two-dimensional in the spot price $S$ and the instantaneous variance $v$). In each case, NNLCI reduces the root-mean-square error (RMSE) of the refined-mesh numerical solution by a factor of approximately 4-12 on test sets, even when the neural network is trained on only a small subset of parameter combinations. These results demonstrate that NNLCI significantly reduces computational requirements for high-dimensional problems in real-time options trading and risk management, offering low training costs and strong generalization ability. |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2608.02778 |
| By: | Pawel Struski; Jakub Swistak; Inez Okulska; Przemyslaw Biecek |
| Abstract: | Large language models (LLMs) are increasingly deployed as economic agents, yet there is little evidence whether LLM agents are suited for participating in market mechanisms designed for humans, and whether these mechanisms deliver desired outcomes when faced with LLM agents. We address this question by replicating seminal economic experiments, replacing human subjects with LLM agents. We place agents in a double auction environment, which is a widely-used market mechanism. We check whether such a market is able to deliver an efficient allocation of resources, thereby testing a novel dimension of alignment of LLM agents -- their compatibility with a fundamental market mechanism. We find that markets populated by LLM agents exhibit slower or no convergence towards market equilibrium, thus providing less efficient allocations than markets populated by humans. We then analyze agents' individual trading decisions and find substantial heterogeneity both across model families and market roles. We also run a lexical analysis of Chain-of-Thought (CoT) traces generated by the agents. We find that the decision to execute a trade rather than continue incrementally adjusting prices is associated with a shift from strategic considerations toward urgency. We publicly release our testing framework, which can be used for future evaluations. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.02580 |
| By: | Rawend Brahem (Central Bank of Tunisia) |
| Abstract: | Forecasting inflation in the presence of changing economic conditions, external shocks, and evolving transmission mechanisms remains an active area of research. This paper applies machine learning (ML) models to forecast headline and core inflation in Tunisia at 1-, 3-, and 6-month horizons, comparing their performance with standard econometric benchmarks within a rolling forecasting framework. Conformal prediction intervals are used to assess forecast uncertainty, while SHAP values serve as an exploratory tool to examine the contribution of explanatory variables to model predictions. The results reveal a clear horizon-dependent pattern, with the predictive gains of ML models increasing at medium and longer horizons. These gains are particularly pronounced at the 6- month horizon, where Support Vector Regression reduces the RMSE by more than 40 percents relative to the best-performing benchmark. Furthermore, SHAP analysis suggests that the relative contribution of predictors varies across forecasting horizons, with inflation persistence playing a more prominent role at short horizons and monetary, external, and commodity price variables becoming more relevant at longer horizons. Overall, these findings suggest that machine learning methods are most valuable as a complement to, rather than a substitute for, traditional forecasting approaches, particularly at longer horizons where nonlinearities become more pronounced. |
| Keywords: | Inflation Forecasting; Machine Learning; Conformal Inference; SHAP Values; Tunisia |
| JEL: | C53 E31 E37 |
| Date: | 2026–09–08 |
| URL: | https://d.repec.org/n?u=RePEc:gii:giihei:heidwp25-2026 |
| By: | Davood Wadi; Yu Ma |
| Abstract: | Search rankings are valuable because human attention is scarce and sequential. Higher-placed alternatives are easier to find, so they are examined and bought more often. Consumers are now delegating search to AI agents that can ingest an entire results page at once. Randomizing the order of one hundred hotel listings across 5, 000 AI agent sessions, we compare four large language models against human field data. AI agents search more deeply than humans and never decline to buy. Position still predicts which listings are inspected, but weakly and non-monotonically: the middle of a results page has the lowest probability of inspection, not the bottom. Position reaches the choice stage for some models and not others, a heterogeneity that tracks neither provider nor capability. All models nonetheless converge on the same undominated listing. For agentic search, the attributes displayed on a results page matter more than placement within it. |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2608.22697 |
| By: | Marcus Buckmann (Bank of England); Galina Potjagailo (Bank of England) |
| Abstract: | This paper discusses how economic theory can be integrated into machine learning (ML) models to enhance their interpretability and applicability for policy analysis. While ML methods offer considerable flexibility and strong predictive performance, they are often criticised for their 'black box' nature and lack of economic transparency. A growing body of research addresses this limitation by introducing structure into ML models − most notably through Block-Additive Models (BAMs) and theory-consistent monotonicity constraints. BAMs group predictors into economically meaningful blocks and impose additivity across blocks, while permitting non-linearities and interactions within them. This architecture enables clear attribution of each block’s contribution to the model’s predictions. Monotonicity constraints further improve interpretability by aligning the model’s directional responses with economic theory, allowing for the separation of opposing effects − such as distinguishing between supply- and demand-driven components of inflation. Empirical evidence shows that these structured ML approaches retain strong predictive performance while yielding economically meaningful narratives. |
| Keywords: | Interpretable machine learning;theory-aligned constraints;macroeconomic analysis |
| JEL: | C10 C14 C53 |
| Date: | 2025–09–26 |
| URL: | https://d.repec.org/n?u=RePEc:boe:boeewp:023266 |
| By: | Marcus Buckmann (Bank of England); Galina Potjagailo (Bank of England); Philip Schnattinger (Bank of England) |
| Abstract: | We propose the Blockwise Boosted Inflation Model (BBIM), a boosted tree framework that decomposes inflation dynamics into predictive components aligned with an open-economy hybrid Phillips curve. Demand and supply contributions are identified by imposing monotonicity constraints, ensuring theory-consistent links between inflation and key indicators. Applied to monthly UK CPI inflation, the model shows that the recent surge has been driven mainly by global supply shocks transmitted through supply chains. We also uncover an L-shaped Phillips curve relationship between inflation and labour market tightness, with tight labour markets amplifying recent inflationary pressures. By contrast, earlier episodes saw non-linearities more strongly tied to broader slack, particularly during recessions. The model further accounts for trend shifts informed by inflation expectations. Short-term household expectations have recently displayed persistent non-linear effects, temporarily raising trend inflation and prolonging inflationary pressures, while longer-term expectations remain anchored. Out-of-sample, the BBIM delivers competitive forecasting performance relative to linear benchmarks and unstructured machine learning methods. Our approach provides a flexible yet interpretable framework that combines economic structure with machine learning for policy-relevant analysis of inflation dynamics. |
| Keywords: | Inflation;Phillips curve;boosted decision trees;machine learning |
| JEL: | E31 E37 C14 C53 |
| Date: | 2025–09–26 |
| URL: | https://d.repec.org/n?u=RePEc:boe:boeewp:023265 |
| By: | Nathan Canen; Ted Enamorado |
| Abstract: | Prediction-based methods, including Large Language Models (LLMs) and other machine learning techniques, are often used to construct measures of political phenomena that are difficult to quantify directly, such as policy positions in manifestos or emotions expressed on social media. In many applications, these prediction-generated measures are used as explanatory variables in regression models, even though they are measured with error. This leads to biased estimates. In this paper, we propose a simple solution to these biases: instrumental variables constructed from multiple measures created on independent splits of the original data. This approach is theoretically valid, easy to implement, and does not require new data. Through simulations, we show that this approach recovers estimates close to the true values, even in relatively small samples, while the standard approach can produce substantial bias in practice. We illustrate the method by revisiting two applications: whether gendered speech affects legislative outcomes in the German Parliament, and whether political risk influences poverty alleviation programs in China. |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2608.02909 |
| By: | Ahmed Asaad; Amr Mohamed; Yang Zhang; Omneya Abdelsalam |
| Abstract: | Large Language Models (LLMs) increasingly use user context such as memory, profiles, and role prompts to personalize their responses. This personalization can affect evidence-based judgment: the same evidence may lead to different conclusions under different user contexts. Finance provides a high-stakes setting to study this problem because decisions often depend on interpreting long and complex documents. We test this using 3, 575 SEC filings across twelve LLMs. We compare persona-conditioned retrieval, neutral retrieval, and memory-framed context to separate the effect of evidence selection from the effect of interpretation. We find that most user-context spillover comes from how models interpret the same evidence under different roles, rather than from retrieving different evidence. We then test two simple mitigation strategies: expressing the same investor mindset as a user profile instead of an assistant role, and separating evidence-based and personalized outputs. Both reduce spillover, but neither removes it completely, and their effectiveness varies substantially across models. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.03218 |
| By: | Raphaël Teixeira (CERAG - Centre d'études et de recherches appliquées à la gestion - UGA - Université Grenoble Alpes, UGA - Université Grenoble Alpes, Grenoble INP - Institut polytechnique de Grenoble - Grenoble Institute of Technology - UGA - Université Grenoble Alpes) |
| Abstract: | Generative Artificial Intelligence (Gen AI) is reshaping professional practices, opening new paradigms of co-creation (Davis, 2013). Human-AI co-creation can be defined as a setting where humans and AI collaborate to create new forms of art, design, and innovation (Vinchon et al., 2023). This research investigates how Human–AI Collaboration impacts creative work. As Artificial Intelligence integration expands, it reshapes productivity, team structures, and creative dynamics across industries (Makarius, 2020). Human–AI Collaboration (HAIC) focuses on cooperative dynamics, shared goals, and team development (Fragiadakis et al., 2024) to understand how agency is built through hybrid intelligence (Sundar, 2020). Managers still lack appropriate tools to evaluate these emerging hybrid collaborations, which vary across sectors (Sauer & Burggräf, 2024). This research targets creative industries and aims to explore how co-creativity emerges as a new collaborative framework. Within the context of agentic AI, this doctoral project intends to design an interaction protocol based on Large Language Models (LLMs) to engage the team's creativity. |
| Keywords: | Multi Agent System, GenAI, Creativity, Human-AI Collaboration |
| Date: | 2026–02–02 |
| URL: | https://d.repec.org/n?u=RePEc:hal:journl:hal-05723566 |
| By: | Jermyn Zhen Yong Bek; Zhuang Qiang Bok; Zhongtian Sun |
| Abstract: | Investment management is a high-stakes domain in which agentic AI systems must do more than generate plausible text. They must retrieve point-in-time data, assemble correct computational inputs, invoke specialized methods, and produce auditable structured outputs. We introduce FinSkillBench, an evaluation suite designed to measure whether language model agents can effectively use financial domain skills to solve investment management tasks. The benchmark spans three domains, portfolio construction, risk management, and fundamental analysis, and includes 12 subtasks with 2, 603 task episodes. Each episode provides point-in-time inputs, hidden ground truth, and a task-specific verifier.We compare three conditions: no skill, curated skill packages consisting of procedural documents and executable components, and self-generated skills in which the agent writes and reuses its own procedures within an episode. Across 9 models and a large-scale evaluation, curated skills consistently improve performance, raising mean scores from 0.366 to 0.528, with the largest gains in portfolio construction and risk management. In contrast, self-generated skills provide little benefit despite higher computational cost. An independent evaluation using a separate agent framework (Hermes Agent, 8 models, 5, 280 episodes total) reproduces the directional pattern across all three domains, with the magnitude of skill effects varying by subtask and harness. These results showthat in investment management agents, access to reliable procedural skills can be as important as model choice, while naive self-generation of skills is often ineffective. We release the benchmark, evaluation tools, curated skill packages, and full trajectories to support further research. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2608.18099 |
| By: | Pattaraphon Kenny Wongchamcharoen; Kris Gulati; Min Min Fong; Abhishek Nagaraj |
| Abstract: | Most LLM benchmarks rank models on their ability to automate work tasks. In practice, however, models are often used to assist other (human or LLM) agents. The question that drives model selection is therefore not only which model produces the best output, but which model most improves the work of another (weaker) agent. We introduce a unified framework that evaluates the capability of models to automate and augment another agent's performance. Across seven economically grounded real-world tasks, an assistant model writes assistance text for a standardized lower-capacity worker model, which produces the deliverable. In automation mode, the assistant produces the output directly. Outputs are scored through blind pairwise comparisons by an LLM judge panel with task-specific rubrics, replicated across ten runs. Rankings across the two regimes are only modestly correlated, and the automation winner loses augmentation on five of seven tasks. Assistance is not reliably positive. The unaided worker outranks every assisted condition on three tasks, and only one model's guidance beats no guidance on average. These results suggest that automation ability is an incomplete proxy for assistance quality, motivating benchmarks that evaluate models according to the roles they play in human-AI and multi-agent systems. |
| JEL: | J24 M15 O33 |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35663 |
| By: | Yingjian Pan; Xiaowei Ding; Kay Giesecke |
| Abstract: | Recent advances in LLM agents enable a new paradigm for asset pricing, which we call Agentic Empirical Asset Pricing (AEAP): systems that autonomously conduct the scientific discovery process itself. We define AEAP and identify its core building blocks. Existing evaluation practices backtest only the outputs (factors or trades), not the autonomous discovery system that produced them. We focus on factor discovery, contributing a reference architecture, a rigorous evaluation standard for discovered factors, and a method for out-of-sample backtesting the discovery system. As a concrete instance of that architecture, we evaluate SEADS against five re-implemented baselines on two US equity panels using this standard: no single metric ranks the systems consistently, motivating evaluation on multiple axes at once. A separate rolling re-execution then asks the complementary question of whether the discovery process itself, not one static output, is reliable. We also report negative findings and limitations that surface further evaluation pitfalls for future AEAP systems. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.00731 |
| By: | André Luduvice; Roberto Pinheiro |
| Abstract: | We model artificial intelligence (AI) as a distinct capital input in a nested CES aggregate production function that extends work done by Krusell et al. (2000) and embed it in a dynamic general equilibrium economy. By bringing the model to the data, we estimate the degree to which AI either substitutes for or complements the other production inputs at the aggregate level. We find that AI is complementary to the high-skill–equipment composite and that the AI weight in production remains small. We then use the model to study how this complementarity shapes the macroeconomic and distributional effects of AI capital accumulation. The estimated model implies different effects of alternative AI shocks. A rise in the AI usage share is contractionary as it increases reliance on a scarce complementary input. A fall in AI prices is expansionary due to the lower cost of accumulating AI capital. A tax on AI capital income raises limited revenue while the AI capital stock remains small, but can finance welfare-improving transfers as the AI price falls. A large-scale universal basic income (UBI) funded jointly by a consumption tax slows AI investment with welfare gains for low-skill workers at the expense of losses for high-skill workers and entrepreneurs. On the measurement side, we construct a quality-adjusted AI price index from hedonic regressions and build a corresponding AI capital stock. |
| JEL: | E22 E25 J24 J31 |
| Date: | 2026–09–04 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedcwq:103765 |
| By: | Andreas Bauer |
| Abstract: | Firms that deploy improving but imperfect AI must decide how much to keep human workers engaged. An agent-based version of the market reproduces the analytical thresholds under noisy beliefs, learning-by-record and worker churn. |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2608.04276 |
| By: | Tetsuya Takaishi |
| Abstract: | Herein, we propose a quantum circuit learning framework for modeling the realized volatility (RV) of Bitcoin and investigate the statistical properties of the predicted time series through multifractal analysis. Unlike conventional GARCH-type models, which require a pre-specified functional form for the volatility process, a parameterized quantum circuit directly approximates the volatility function from empirical data, eliminating the need for explicit model selection. Using five-minute Bitcoin price data, we construct daily RV, train a single-qubit parameterized quantum circuit, and generate a long synthetic time series from the optimized quantum circuit. Multifractal Detrended Fluctuation Analysis is applied to calculate the generalized Hurst exponent $h(q)$, the singularity spectrum $f(\alpha)$, and the multifractal scaling exponent $\tau(q)$. The predicted return series exhibits $h(2)\approx 0.5$, consistent with near-random dynamics, and both the predicted and the empirical return series display multifractality that partially persists after random shuffling. The increment series of RV shows pronounced anti-persistence with $h(2)\approx 0.05$--$0.1$, consistent with the rough volatility hypothesis. These results demonstrate that a simple single-qubit parameterized quantum circuit captures qualitatively some observed properties in Bitcoin volatility dynamics. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.04569 |