Neural Networks for Observable Market Data Validation

Observable Market Data is critical for effective valuation of trades for risk management purposes within the investment bank. The valuation process requires the existence of good quality data day by day, and dating back into the mid-2000’s. Not all assets have highly liquid data available either historically or at present, and there is significant interest within the industry in building models to both predict missing data and qualify available data. Historically this process has been highly manual, and due to the volume of data statistical methods are used to identify potentially suspect data. The use of these simplistic methods to gate incoming data results in known blind spots and false positives. This project seeks to use deep neural networks to tackle these two tasks : 1) flagging suspect data efficiently, 2) generating quality data that can be used to improve modeling where real data is unavailable.

Faculty Supervisor:

Kirill Serkh;Vardan Papyan

Student:

Partner:

CIBC

Discipline:

Computer science

Sector:

Finance and Insurance

University:

University of Toronto

Program: