Applying Simulation to the Problem of Detecting Financial Fraud

Abstract: This thesis introduces a financial simulation model covering tworelated financial domains: Mobile Payments and Retail Stores systems.The problem we address in these domains is different types offraud. We limit ourselves to isolated cases of relativelystraightforward fraud. However, in this thesis the ultimate aim is to introduce our approach towards the use of computer simulation for fraud detection and its applications in financial domains. Fraud is an important problemthat impact the whole economy. Currently, there is a general lack ofpublic research into the detection of fraud. One important reason isthe lack of transaction data which is often sensitive. To address thisproblem we present a Mobile Money Simulator (PaySim) and RetailStore Simulator (RetSim), which allow us to generate synthetictransactional data that contains both: normal customer behaviour and fraudulent behaviour.These simulations are multi agent based simulations and were calibrated using real data from financial transactions. Wedeveloped agents that represent the clients and merchants in PaySim andcustomers and salesmen in RetSim. The normal behaviour wasbased on behaviour observed in data from the field, and is codified inthe agents as rules of transactions and interaction between clients and merchants,or customers and salesmen. Some of these agents were intentionallydesigned to act fraudulently, based on observed patterns of realfraud. We introduced known signatures of fraud in our model andsimulations to test and evaluate our fraud detection methods. Theresulting behaviour of the agents generate a synthetic log of alltransactions as a result of the simulation. This synthetic data can beused to further advance fraud detection research, without leakingsensitive information about the underlying data or breaking any non-disclose agreements.Using statistics and social network analysis (SNA) on real data wecalibrated the relations between our agents and generaterealistic synthetic data sets that were verified against the domain and validated statistically against the original source.We then used the simulation tools to model common fraud scenariosto ascertain exactly how effective are fraud techniques such as the simplest form of statisticalthreshold detection, which is perhaps the most common in use. The preliminary resultsshow that threshold detection is effective enough at keeping fraudlosses at a set level, that there seems to be little economic room forimproved fraud detection techniques.We also implemented other applications for the simulator tools such as the set up of a triage model and the measure of cost of fraud. This showed to be an important help for managers that aim to prioritize the fraud detection and want to know how much they should invest in fraud to keep the loses below a desired limit according to different experimented and expected scenarios of fraud.