Introduction to Financial Fraud

Basic introdution (according to reports)

  • types of crime (according to PwC)
    • accounting/financial statement fraud
    • anti-competition/antitrust law infringement
    • bribery and corruption
    • customer fraud
    • cybercrime
    • deceptive business practices
    • human resources fraud
    • insider/unauthorized trading
    • intellectual property (IP) theft
    • money laundering and sanctions
    • procurement fraud
  • interesting phenomenon
    • approxiamtely 50% of companies experienced fraud in the past 2 years
    • but only 56% of companies conducted investigations into their worst incidents

Fraud

  • definition
    • no formal definition
    • English dictionary definition
      • the crime of getting money by deceiving people
    • legal dictionary definition
      • the intentional use of deceit, a trick or some dishonest means to deprive another of his/her/its money, property or a legal right
  • types of financial frauds
    • no single definition
    • categorization (by FBI)
      • bank fraud
        • credit card fraud
          • application fraud
            • obtain new credit cards using false personal information
            • spending as much as poosible in a short time
          • behavioral fraud
            • stole details of credit cards (not physical card), using on a “cardholder not present” basis
            • most common type
        • mortgage fraud
        • money laundering
          • make illegally obtained gain and proceeds seemingly legal
          • AML (Anti-Money Laundering) compliance is mandatory requirements for running business in the financial sector
      • corporate fraud
        • financial statement fraud
          • represent fake financial condition of a company
          • deveive or mislead to the users of the financial statements
          • e.g. Enron
        • securities and commodities fraud
      • insurance fraud
        • any types of insurance
          • automobile insurance fraud (by FBI)
          • health care fraud (by FBI)
        • from the issuer (seller)
          • selling policies from non-existent companies
          • failing to submit premiums
          • churning policies to create more commissions
        • from the buyer
          • exaggerated claims
          • falsified medical history
          • post-dated policies
          • faked death
          • faked damage
    • other types
      • identity theft
        • replace someone to make transactions or purchases
        • stealing lists of customer information
      • tax evasion
        • underpay taxes
      • securities fraud
        • a false statement about a company or company’s valued stock
        • others make decisions based on that lie
  • fraud triangle (actually, a circle)
    • pressure
      • from financial, social, or any other nature
      • it cannot be resolved or relieved in an authorized manner
    • opportunity
      • opportunity exists to resolve or relieve the experienced pressure in an unauthorized but concealed or hidden manner
    • rationalization
      • pyschological mechanism, explain why fraudsters do not refrain
      • think conduct as acceptable

The Fraud Cycle

Overview

Details

  • fraud investigation (current practice)
    • a human expert investigates suspicious, flagged cases
  • fraud confirmation (current practice)
    • determine true fraud label
    • after this, 2 measures will be taken
      • corrective measures
        • resolve the fraud and correct the wrongful consequences -> fix
      • preventive measures
        • preventing future fraud by the same fraudster or others -> prevent
  • automated detection algorithm
    • newly detected cases should be addded to the database of historical fraud cases -> optimize the model
  • fraud detection
    • apply detection model on new, unseen observations
    • assign a fraud risk to every observation
  • fraud prevention
    • prevent fraud in the future
    • detect or predict fraud before it happens
  • fraud detection model
    • frequency of re-training the model depends on
      • volatility of the fraud behaviour
      • detection power of the current model
      • amount of (similar) confirmed cases already available in the database
      • rate at which new cases are being confirmed
      • required effort to retrain the model

Fraud Detection and Prevention

Overview

  • fraud detection (post approach)
    • recognize or discover fradulent activities
    • traditionally, expert-based approach
      • depends on a human’s professional knowledge to build a number of rule-based engines (IF…, THEN…)
      • disadvantages
        • expensive to build -> manual input
        • difficult to maintain -> manually and regularly updated
        • slow response time to new fraud cases -> expert needs to study new fraud cases
        • prone to human error -> relies heavily on human expert
    • reasons for using data driven fraud detection
      • precision
        • process massive volumes of data
        • more efficient than human eyes
      • operational efficiency
        • process a huge amount of cases in a short time
      • cost efficiency
        • compared with expert-based system, better cost efficiency and ROI
  • fraud prevention (ante approach)
    • avoid or reduce fraud
    • complementary to fraud detection, not independent of fraud detection

Fraud analytics process model

  • overview
    • is not a linear process but a cyclic process
    • one may go over to the previous steps during the process
  • pre-processing
    • identify business problem is the most important task in this part
  • fraud analytics
    • techniques used
      • expert-based techniques
      • statistical techniques
      • AI/data mining/machine learning techniques
  • post-processing
    • key characteristics of a good fraud analytics model
      • statistical accuracy
        • accuracy is not the only parameter, e.g. capacity or capability of the model
        • statistically significant, i.e. the patterns found is not the consequence of coincidence
      • interpretability
        • black-box models are acceptable
      • operational efficiency
        • real-time evaluation
        • re-estimate the model when necessary
      • economical cost
        • tech and human cost of building models
      • regulatory compliancy
        • should comply with all regulation and legislation, e.g. privacy laws

Financial fraud analytics

  • red flag
    • anomality or variation from predictable patterns of normal behavious
    • exist in
      • data
      • documents
      • internal controls
      • behaviors
      • public records
      • digital footprints, etc.
    • examples
      • tax evasion fraud
        • an identical financial statement, copy nonfraudulent companies
        • name of an accountant is unique, this accountant may not exist
      • credit card fraud
        • a small payment followed by a large payment immediately after, test the card is still active
        • regular rather small payments, may avoid getting noticed
  • good practice
    • plan before acting
    • well-defined scope for fraud analytics
    • good understanding of fraud data
      • distinguish red flags and error in data
      • identify false positives and fasle negatives
      • e.g. someone uses credit card to purchase frequently in Dec, which may be not fraud due to Christmas
    • try not to jump to conclusions without correct and logical linkage to evidence found