Threshold assessment

Which of the following scenarios are statistical outlier scenario?

• SAS10033 Politically Exposed Person (PEP) Indicator

  A risk factor, triggered if customer is a a PEP

• SAS10006 Large Cash Transactions

  Alert if aggregated cash transactions in past 14 days is greater than $100,000

• SAS10021 High Velocity Wire-in

  Alert if in past 14 days, aggregate deposit through wire is large, and most of this amount is withdrawn within the period.

• All of the above

Which of the following is true about ATL and BTL?

• Above-the-line test can reveal new true positive alert
• Below-the-line test involve human decision on whether alert is a true positive
• Both ATL and BTL can be done automatically

Which of the following is NOT the tuning Objective?

• Increase Productivity of Alerts
• Increase accuracy of Alerts
• Increase amount of Alerts
• Meet Compliance requirement

Which is the rate to measure Type I error?

• Precision Rate
• Recall Rate
• Accuracy Rate
• Return Rate

Why is it rare to define Productive Alerts as the "Actual submissions of regulatory report"?

• Disclosure of information about submitted alerts is not allowed
• Alerts that lead to the actual submission may be so small that it becomes statistically irrelevant
• There is a big time lag between actual submission time and alert occurrence time since investigation takes time
• All of the above

Which of the following is/are Common Scenario Parameters?

• Aggregate Amount
• Aggregate Count
• Percentage Rate
• All of the above

What is a benefit of data driven method for customer segmentation?

• Implementation effort is low
• It always give more accurate result
• It takes out bias of domain expert
• Required data may not exist

What distribution does a parameter variable (numeric variable that is compared against threshold for alert condition) usually follows?

• Binomial distribution
• Normal distribution
• Uniform distribution
• Log normal distribution

If a parameter variable (numeric variable that is compared against threshold for alert condition) distribution has multiple peaks, which of the following best describes the situation?

• The contributing customers may contain multiple heterogenous groups that may better be further split for customer segments
• A point between the peaks is a good candidate for optimal threshold
• There is data quality issue that should first be tackled before analysis
• The peak with highest value is the threshold we are looking for

Which of the following is not considered in threshold setting?

• Risk level of the customer
• Customer segmentation
• Alert investigation process
• Transaction type