Large language models predict household economic behavior and policy responses
HouseholdBench: Evaluating Large Language Models as Predictors of Household Economic Behavior
Computation and Language
Summary
Understanding how families make financial choices is important but hard to model. The authors created HouseholdBench, a new test combining multiple surveys and prediction tasks to see how well large language models can guess what households will do with money, work, and housing under different conditions. They found that many language models predict better than simple guesses and can even anticipate how people change when policies shift. Though machine learning trees often do better, some language models approach that level with extra training and combining multiple predictions.
What this means in practice
- •For economic policy analysts: Use language models to predict how households adjust behaviors under economic policy changes using a diverse set of survey data.
- •For financial service developers: Implement fine-tuned open-weight language models to enhance household financial behavior forecasts for consumer-facing applications.$Commercial implications: Allows development of personalized financial planning tools that better predict client economic responses using scalable, open models.
Authors
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: https://jn-huang.github.io/householdbench