This paper investigates whether large language models (LLMs), specifically GPT-4, can perform financial statement analysis as effectively as professional human analysts. The study provides standardized and anonymized financial statements to GPT-4 and asks the model to predict future earnings direction. The results show that GPT-4 outperforms human analysts in predicting earnings changes, particularly in situations where analysts struggle. The model's predictions are comparable to those of a state-of-the-art artificial neural network (ANN) model. The study also finds that GPT-4 generates useful narrative insights about a company's future performance, which are not derived from its training data but from its ability to analyze financial numbers. Trading strategies based on GPT-4 predictions yield higher Sharpe ratios and alphas than strategies based on other models. The research suggests that LLMs may play a central role in financial decision-making. The study uses the Compustat universe and intersects it with the IBES universe, covering a large sample of firm-year observations. The results indicate that GPT-4's predictions are accurate and valuable, even without narrative context. The study also finds that GPT-4's predictions are complementary to those of human analysts, with each providing unique insights. The findings suggest that LLMs can complement human analysts in financial analysis, offering valuable insights and improving prediction accuracy. The study highlights the potential of LLMs in financial markets, particularly in tasks that require numerical analysis and judgment. The results suggest that LLMs may take a more central role in decision-making than previously thought. The study also contributes to the literature on fundamental analysis, showing that LLMs can generate state-of-the-art inferences about the direction of earnings, outperforming financial analysts and prior models. The study finds that GPT-4's predictions are accurate and valuable, even without narrative context, and that the model's ability to generate narrative insights is a key factor in its performance. The study also finds that GPT-4's predictions are complementary to those of human analysts, with each providing unique insights. The results suggest that LLMs can complement human analysts in financial analysis, offering valuable insights and improving prediction accuracy. The study highlights the potential of LLMs in financial markets, particularly in tasks that require numerical analysis and judgment. The results suggest that LLMs may take a more central role in decision-making than previously thought.This paper investigates whether large language models (LLMs), specifically GPT-4, can perform financial statement analysis as effectively as professional human analysts. The study provides standardized and anonymized financial statements to GPT-4 and asks the model to predict future earnings direction. The results show that GPT-4 outperforms human analysts in predicting earnings changes, particularly in situations where analysts struggle. The model's predictions are comparable to those of a state-of-the-art artificial neural network (ANN) model. The study also finds that GPT-4 generates useful narrative insights about a company's future performance, which are not derived from its training data but from its ability to analyze financial numbers. Trading strategies based on GPT-4 predictions yield higher Sharpe ratios and alphas than strategies based on other models. The research suggests that LLMs may play a central role in financial decision-making. The study uses the Compustat universe and intersects it with the IBES universe, covering a large sample of firm-year observations. The results indicate that GPT-4's predictions are accurate and valuable, even without narrative context. The study also finds that GPT-4's predictions are complementary to those of human analysts, with each providing unique insights. The findings suggest that LLMs can complement human analysts in financial analysis, offering valuable insights and improving prediction accuracy. The study highlights the potential of LLMs in financial markets, particularly in tasks that require numerical analysis and judgment. The results suggest that LLMs may take a more central role in decision-making than previously thought. The study also contributes to the literature on fundamental analysis, showing that LLMs can generate state-of-the-art inferences about the direction of earnings, outperforming financial analysts and prior models. The study finds that GPT-4's predictions are accurate and valuable, even without narrative context, and that the model's ability to generate narrative insights is a key factor in its performance. The study also finds that GPT-4's predictions are complementary to those of human analysts, with each providing unique insights. The results suggest that LLMs can complement human analysts in financial analysis, offering valuable insights and improving prediction accuracy. The study highlights the potential of LLMs in financial markets, particularly in tasks that require numerical analysis and judgment. The results suggest that LLMs may take a more central role in decision-making than previously thought.