Deficiency of Large Language Models in Finance: An Empirical Examination of Hallucination