Overview¶
Transaction Monitoring Systems generate alerts based on business rules that utilize specific parameter values and thresholds, which are criteria for what qualifies as an alert.These parameter and threshold values are tuned following a predominantly manual process by investigation teams.
- Baseline Thresholds
- Sample Sizes
- ATL/BTL Sampling
- Parameter Analysis
- Threshold Assessment
Business Requirements¶
The below methodology needs to be incorporated with the existing ones in the automation tool:
IQR Methodology¶
- IQR is a statistical methodology that helps identify mild and extreme outliers for the population distribution.
- It is defined as the difference between the first and the third quartiles (P75-P25).
- It can be used as a measure of how spread out the values are and hence gives an estimate of where the outliers lie.
- Mild outliers = P75 value + (IQR value * 1.5)
- Extreme outliers = P75 value + (IQR value * 3)
Note
A SQL script that generates values for parameters from 50th percentile to 99th percentile helps with threshold analysis.
IQR - Min Agg Trans Amt | Median | IQR | P25 | P75 | Mild Outlier | Extreme Outlier |
---|---|---|---|---|---|---|
Total Credit Amount | $ - | $204,080.80 | $ - | $204,080.80 | $510201.90 | $816323.00 |
Total Debit Amount | $100,416.00 | $100,416.00 | $ - | $222,513.00 | $556282.50 | $890052.00 |
Alerts with parameter values higher than mild and/or extreme outlier values can be analysed along with other methodologies to provide recommendations reset thresholds. The ultimate decision of setting thresholds depends on the analyst
Scenario Tuning Components¶
The below requirements are applicable to all methodologies (Percentile Select, Percentile Jump, Productivity Rate and IQR) and need to be automated:
- Baseline Thresholds: Select initial thresholds (percentiles) for different parameters (such as transaction amount, transaction count, etc.) that determine generation of alerts
- Sample Sizes: Select statistically significant number of alerts that are a fair representation of the population by using hypergeometric distribution
- Confidence Level: 95%
- Error Rate: 5%
- Population size: 20,000
- Response Distribution: 50%
Sampling example:
Source System | CINQ |
---|---|
Source System Code | 111 |
Inst./ Channel Type | Wires |
Daily Avg. Txn. Volume | 140,055 |
Sample Size | 384 |
http://www.raosoft.som/samplesize.html
- Parameter Analysis: Based on the rules logic, select the parameters that need to be tuned
- ATL/BTL Testing: Generate BTL alerts and ATL alerts for the selected Baseline Thresholds and Sample Sizes
For instance, with a Baseline Threshold of 80th percentile, BTL analysis considers 75th percentile (below the 80th percentile) and generates alerts. ATL analysis considers 85th percentile (above the 80th percentile) and generates alerts. BTL/ATL analysis takes (+/-5%, +/-10%, etc. based on feedback received by the investigation team for each of the samples)
- False positives
- Number of alerts that led to creation of SARS or FIs
Based on their analysis, thresholds for parameters are reset manually
Note
The sampling process still requires feedback from investigative teams and decision making.
Abbreviations | Description |
---|---|
IQR | Inter Quartile Range |
BTL | Below the Line |
ATL | Above the Line |
SAR | Suspicious Activity Report |
FI | Field Inquiries |