How to reduce prediction errors when doing predictions on long-term stock performance.
How to handle data biases and missing values associated with 20-30 years of multi-stock multi-frequency data.
JMI Implementation
Assessed client’s Time Series-based model and determined that biases and missing values were not handled optimally - disrupting prediction engines.
Consolidated uncontrollable number of objective functions containing multiple lagging variables to much simpler objective functions - reducing model complexity and improving efficiency of portfolio returns using dynamic optimization techniques.
Leveraged proprietary AI runtime to identify and eliminate survivorship, look-ahead, optimization and other biases and helped traders to transform the patterns.
Using our trained and tested models, JMI helped clients to reduce overfitting because of fewer variables utilized - model performance stabilization.
This resulted in reduction of false signals generated by earlier system due to missing values and data biases.
Results
Prediction error reduction by 15% for 9-quarter and 13-quarter forecasts.
Client can define prediction windows of any duration, not worrying about missing values and bias.
Client has experienced an improvement of 14% in predicting risks that improved overall performance.
