Quantifying Soft Information, Mortgage Market Efficiency & Asset Pricing Implications

Abstract

I provide a novel framework for machine learning models to ingest quantified soft information during the life of a loan, using cutting-edge natural language processing techniques on salient unstructured text. This soft information, from servicer call transcripts, is not restricted to mere positive/negative sentiments and provides efficiency and alleviates the information asymmetry between the lender (and/or issuer) and the borrower. Proprietary servicer comments are hardly accessible and offer the soft in-formation for real-time delinquency status of the mortgages. I investigate whether the special servicer invoked by the investor can utilize the valuable comments from the master servicer. The time-varying soft information about the borrower’s financial condition, health of the loan and the property condition from these master servicer comments renders the predictive power and has asset pricing implications. Given this valuable information, the special servicer may choose to use this information, as I anecdotally see with several private equity investors. The well-known unresolved conflict of interest between the master and special servicers can be resolved, thereby reducing moral hazard and increasing efficiency and transparency.