This thesis investigates whether Large Language Models (LLMs) can serve assynthetic participants in behavioral economics experiments, focusing on temporal framing effects in savings decisions. Using a prior human-subject study as a bench- mark, synthetic agents were constructed to replicate and extend the finding that daily savings framings increase participation more than monthly framings. Weak generalization matched the original sample’s traits, while strong generalization ap- plied the design to gig economy workers. The results partially replicate original findings and show strong support for framing effects in new populations. This thesis demonstrates the potential of LLM-generated agents to support behavioral research in scalable, cost-effective, and ethically constrained settings.