For decades, investor have relied on intuition, a limitedset of financial ratios, and traditional regression analyses to predict future earnings changes. Are detailed financial data even useful? We present empirical evidence that combining detailed financial data with machine learning models significantly enhances predictive accuracy. By using eXtensible Business Reporting Language (XBRL) to extract over 13,000 predictors from financial statements and applying two state-of-the-art machine learning models, we can predict firms' future earnings changes more accurately than traditional methods. Our findings show that this improved performance is due to both the detailed financial data and the capability of machine learning models to handle complex relationships between predictors. Moreover, a trading strategy formed based on the machine learning models yields positive returns. Therefore, our findings suggest that detailed financial data and machine learning are useful in predicting the direction of earnings changes and provides an actionable investment strategy.