Random Walk Based Trade Reference Computation for Personal Credit Scoring

Abstract

Personal credit scoring is a fundamental problemin Finance. It has a lot of emerging applications includingCredit Card, Mortgage Loan and Automobile Credit, etc. Dueto its importance, a lot of personal credit scoring methods havebeen proposed. Among many other aspects of a client, tradereference is a crucial aspect for an effective credit scoring. Ina trade reference, the referee provides a score to describe thecredit of the client with respect to the business between theclient and the referee. Most existing methods assume that thetrade reference is trustable. However, the trade referee mayconspire with the client and the trade reference score becomesuntrustable in practice. Therefore, in this paper, we propose atrade referee rank to capture both the reputation of the tradereferee and the trade reference score provided by the referee. The accuracy of using our trade reference rank is about 40%higher than that of the existing methods. We propose a randomwalk based method on a client reference graph to compute thetrade reference rank of the clients. Our extensive experimentsconfirm the effectiveness and the efficiency of our proposed techniques.