|
on Artificial Intelligence |
| By: | Annie Liang |
| Abstract: | Machine learning algorithms can now outperform classic economic models in predicting quantities ranging from bargaining outcomes, to choice under uncertainty, to an individual's future jobs and wages. Yet this predictive accuracy comes at a cost: most machine learning algorithms function as black boxes, offering little insight into \emph{why} outcomes occur. This article asks whether machine learning can guide the development of new economic theories. Economic models serve an important purpose beyond prediction -- they uncover the general mechanisms behind observed behaviors. A model that identifies the causal pathways of economic development is more valuable than one that merely predicts which countries will escape poverty, because it enables policymakers to encourage that development in countries where it might not have happened otherwise. Similarly, a model that predicts imperfectly across many domains can be more valuable than one that is highly accurate in a specific domain, since the former allows insights and data obtained from one setting to inform decisions and policy in another. Applying machine learning algorithms off-the-shelf is unlikely to yield such models. But recent work shows that, when reconceived with the aims of an economic modeler in mind, machine learning methods can improve both prediction and understanding. These approaches range from adversarially training algorithms to expose the limits of existing models, to imposing economic theory as a constraint on algorithmic search. Advances in large language models complement these strategies and open new research directions. |
| Date: | 2025–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2508.19136 |
| By: | Maxim Chupilkin |
| Abstract: | How does AI think about economic policy? While the use of large language models (LLMs) in economics is growing exponentially, their assumptions on economic issues remain a black box. This paper uses a conjoint experiment to tease out the main factors influencing LLMs' evaluation of economic policy. It finds that LLMs are most sensitive to unemployment, inequality, financial stability, and environmental harm and less sensitive to traditional macroeconomic concerns such as economic growth, inflation, and government debt. The results are remarkably consistent across scenarios and across models. |
| Date: | 2025–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2507.15771 |
| By: | Benjamin G. Hyman; Benjamin Lahey; Karen Ni; Laura Pilossoph |
| Abstract: | We document the extent to which workers in AI-exposed occupations can successfully retrain for AI-intensive work. We assemble a new workforce development dataset spanning over 1.6 million job training participation spells from all US Workforce Investment and Opportunity Act programs from 2012–2023 linked with occupational measures of AI exposure. Using earnings records observed before and after training, we compare high AI exposure trainees to a matched sample of similar workers who only received job search assistance. We find that AI-exposed workers have high earnings returns from training that are only 25% lower than the returns for low AI exposure workers. However, training participants who target AI-intensive occupations face a penalty for doing so, with 29% lower returns than AI-exposed workers pursuing more general training. We estimate that between 25% to 40% of occupations are “AI retrainable” as measured by its workers receiving higher pay for moving to more AI-intensive occupations—a large magnitude given the relatively low-income sample of displaced workers. Positive earnings returns in all groups are driven by the most recent years when labor markets were tightest, suggesting training programs may have stronger signal value when firms reach deeper into the skill market. |
| JEL: | E0 E2 J6 |
| Date: | 2025–08 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:34174 |
| By: | Lukas B. Freund; Lukas F. Mann |
| Abstract: | Who will gain and who will lose as AI automates tasks? While much of the discourse focuses on job displacement, we show that job transformation—a shift in the task content of jobs—creates large and heterogeneous earnings effects. We develop a quantitative, task-based model where occupations bundle multiple tasks and workers possessing heterogeneous portfolios of task-specific skills select into occupations by comparative advantage. Automation shifts the relative importance of tasks within each occupation, inducing wage effects that we characterize analytically. To quantify these effects, we measure the task content of jobs using natural language processing, estimate the distribution of task-specific skills, and exploit mappings to prominent automation exposure measures to identify task-specific automation shocks. We apply the framework to analyze automation by large language models (LLMs). Within highly exposed occupations, like office and administrative roles, workers specialized in information-processing tasks leave and suffer wage losses. By contrast, those specialized in customer-facing and coordination tasks stay and experience wage gains as work rebalances toward their strengths. Our findings challenge the common assumption that automation exposure equates to wage losses. |
| Keywords: | AI, labor markets, inequality, skills, technological change |
| JEL: | J01 E00 J23 J24 O33 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_12072 |
| By: | Mayuree Santhadkitjakarn (Chulalongkorn University, Bangkok, Thailand); Somtip Watanapongvanich (Chulalongkorn University, Bangkok, Thailand) |
| Abstract: | This research explores the effects of artificial intelligence (AI) on the labor market across various regions of China, using provincial-level data from 2006 to 2022. To address issues of reverse causality and ensure analytical accuracy, a fixed-effects distributed lag model is employed. The study investigates three key dimensions: total employment, wages, and employment across different skill levels. COVID-19 indicators are included to account for policy changes and economic fluctuations. The results reveal a relationship between the rising adoption of AI and employment opportunities for low- to medium-skilled workers; however, high-skill occupations appear less affected. Evidence also suggests that the integration of AI technologies influences wages, consistent with the assumptions of skill-biased technological change (SBTC). These findings provide empirical confirmation of the disruptive impact of AI on the Chinese labor market, offering insight into how technological advancement affects employment in emerging economies. The study offers guidance for policymakers and corporate leaders addressing labor challenges in an AI-driven economy. |
| Keywords: | AI, Employment Structure, Total Employment, Wage |
| Date: | 2025–04 |
| URL: | https://d.repec.org/n?u=RePEc:smo:raiswp:0506 |
| By: | Andrew Blair-Stanek; Nils Holzenberger; Benjamin Van Durme |
| Abstract: | We investigate whether large language models can discover and analyze U.S. tax-minimization strategies. This real-world domain challenges even seasoned human experts, and progress can reduce tax revenue lost from well-advised, wealthy taxpayers. We evaluate the most advanced LLMs on their ability to (1) interpret and verify tax strategies, (2) fill in gaps in partially specified strategies, and (3) generate complete, end-to-end strategies from scratch. This domain should be of particular interest to the LLM reasoning community: unlike synthetic challenge problems or scientific reasoning tasks, U.S. tax law involves navigating hundreds of thousands of pages of statutes, case law, and administrative guidance, all updated regularly. Notably, LLM-based reasoning identified an entirely novel tax strategy, highlighting these models' potential to revolutionize tax agencies' fight against tax abuse. |
| Date: | 2025–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2508.20097 |
| By: | Hui Chen; Antoine Didisheim; Luciano Somoza; Hanqing Tian |
| Abstract: | Emerging techniques in computer science make it possible to "brain scan" large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance. We also show that models can be steered to be more or less risk-averse, optimistic, or pessimistic, which allows researchers to correct or simulate biases. The method is transparent, lightweight, and replicable for empirical research in the social sciences. |
| Date: | 2025–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2508.21285 |
| By: | Fabrizio Dimino; Krati Saxena; Bhaskarjit Sarmah; Stefano Pasquali |
| Abstract: | The growing adoption of large language models (LLMs) in finance exposes high-stakes decision-making to subtle, underexamined positional biases. The complexity and opacity of modern model architectures compound this risk. We present the first unified framework and benchmark that not only detects and quantifies positional bias in binary financial decisions but also pinpoints its mechanistic origins within open-source Qwen2.5-instruct models (1.5B--14B). Our empirical analysis covers a novel, finance-authentic dataset revealing that positional bias is pervasive, scale-sensitive, and prone to resurfacing under nuanced prompt designs and investment scenarios, with recency and primacy effects revealing new vulnerabilities in risk-laden contexts. Through transparent mechanistic interpretability, we map how and where bias emerges and propagates within the models to deliver actionable, generalizable insights across prompt types and scales. By bridging domain-specific audit with model interpretability, our work provides a new methodological standard for both rigorous bias diagnosis and practical mitigation, establishing essential guidance for responsible and trustworthy deployment of LLMs in financial systems. |
| Date: | 2025–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2508.18427 |
| By: | Haojie Liu; Zihan Lin; Randall R. Rojas |
| Abstract: | This study integrates real-time sentiment analysis from financial news, GPT-2 and FinBERT, with technical indicators and time-series models like ARIMA and ETS to optimize S&P 500 trading strategies. By merging sentiment data with momentum and trend-based metrics, including a benchmark buy-and-hold and sentiment-based approach, is evaluated through assets values and returns. Results show that combining sentiment-driven insights with traditional models improves trading performance, offering a more dynamic approach to stock trading that adapts to market changes in volatile environments. |
| Date: | 2025–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2507.09739 |