Are missing values important for earnings forecast? a machine learning perspective.

Area of Science:

• Financial econometrics
• Machine learning applications
• Data imputation techniques

Background:

• Analysts' forecasts are crucial for estimating firm earnings but often contain missing values.
• Existing methods struggle to fully leverage incomplete forecast data, impacting prediction accuracy.

Purpose of the Study:

• To apply machine learning for imputing missing analyst forecast data.
• To predict firm future earnings using both imputed and observed forecasts.
• To evaluate and compare different imputation methods for financial forecasting.

Main Methods:

• Utilized machine learning techniques, including matrix factorization (MF), to impute missing values in individual analyst forecasts.
• Developed a stochastic gradient descent-based coupled matrix factorization (CMF) model integrating multiple datasets.
• Evaluated imputation performance and its impact on earnings forecast accuracy.

Main Results:

• Imputing missing values reduced forecast error by 41% compared to using the mean forecast.
• Matrix factorization (MF) demonstrated consistent out-performance across various evaluation metrics and firms.
• Coupled matrix factorization (CMF) further reduced earnings forecast error by an additional 19% compared to MF using a single dataset.

Conclusions:

• Machine learning imputation significantly enhances the utility of analyst forecasts for earnings prediction.
• Matrix factorization is a robust method for handling missing forecast data.
• Coupled matrix factorization offers superior imputation quality and forecast accuracy by leveraging multiple data sources.