Objective

• A leading US-based quantitative systematic hedge fund wanted to improve data quality through exploration of imputation of missing values for financial statements using Machine Learning techniques with constraint to maintain accounting consistency between the statements
• The client was evaluating different vendors for the purpose of this pilot project
• Tools Used: R
  

CRISIL's solution

• The client provided large data extract from a single market vendor for nearly 9000 US companies. Set up a team of Machine Learning and Accounting experts to meet this unique requirement of accounting identities and relationships
• Carefully assessed missing data and client requirements. Proposed the MICE approach to impute missing values. 
• Used Expectation Maximization algorithm with statistical techniques like auto-regression and PMM to arrive at final results
• Client appreciated the MICE approach and was impressed with the accuracy of the results generated compared to the suggested approach and results provided by other vendors
• The model was validated using different sub samples of data, using in-sample and out-sample techniques, validation of output against actual reported results using small sample of companies for key line items
• Model performance was measured using Mean Absolute Deviation and RMSE for benchmarking against using previous year values, pmm as imputation methodology
  

Client impact

• Robust and ML based solution to strengthen data cleaning, handling missing/outlier values for financial statements 
• Better quality data relevant for any fundamental analysis of companies across equities/credit portfolio
  

Questions

Looking for high-end research and risk services? Reach out to us at: