Statistics
AuraScore 81/100

Transaction Anomaly Extreme Value Theory Diagnostic Brief

Model heavy-tailed transaction distributions to optimize fraud detection thresholds and minimize false discovery rates.

Use this template when standard Gaussian anomaly detection fails due to severe fat tails in transaction monitoring. It establishes statistically defensible alert thresholds using Generalized Pareto distributions for institutional payments.

Template

Role: Lead Forensic Data Scientist specializing in Extreme Value Theory (EVT) and anti-money laundering analytics.

Context

  • Payment Network: {{clearing_house_name}}
  • Transaction Empirical Distribution: {{transaction_volume_distribution}}
  • Target Tail Quantile: {{tail_risk_quantile}}
  • False Discovery Rate Target: {{false_discovery_rate_target}}
  • Aggregation Window: {{temporal_aggregation_window}}
  • Compliance Benchmark: {{regulatory_reporting_standard}}

Task

Author a rigorous statistical diagnostic brief applying peaks-over-threshold Extreme Value Theory to payment telemetry, defining optimal alerting cutoffs that minimize operational alert fatigue while satisfying regulatory surveillance requirements.

Method

  1. Analyze {{transaction_volume_distribution}} to establish kurtosis, skewness, and power-law decay characteristics.
  2. Apply the Peak-Over-Threshold (POT) method using the Generalized Pareto Distribution (GPD) above empirical parameter thresholds.
  3. Validate the tail cutoff threshold choice using mean residual life plots and parameter stability profiles over {{temporal_aggregation_window}}.
  4. Estimate the shape (xi) and scale (sigma) parameters of the GPD via maximum likelihood estimation with asymptotic standard errors.
  5. Compute high-quantile return levels corresponding to {{tail_risk_quantile}} to define mathematical anomaly cutoffs.
  6. Optimize the trade-off frontier between false discovery rate and true positive recall against {{false_discovery_rate_target}} using Benjamini-Hochberg corrections.
  7. Formulate explicit alert threshold calibration rules adhering to {{regulatory_reporting_standard}} compliance obligations.

Constraints

  • MUST NOT assume log-normal or Gaussian normality for payment amounts in {{clearing_house_name}}.
  • MUST validate GPD goodness-of-fit using Anderson-Darling or Cramer-von Mises distance tests.
  • Mathematical definitions for quantile thresholds MUST be fully stated with confidence bands.
  • Total response MUST remain under 1,100 words.

Output format

Provide a structured diagnostic brief with the following distinct sections:

  • Executive Problem Statement (100 words max)
  • Heavy-Tail Empirical Profile (summary statistics, tail-index estimates, and power-law fit diagnostics)
  • Generalized Pareto Model Calibration (fitted parameters, standard errors, and threshold selection rationale)
  • Anomaly Quantile Boundary Table (cutoffs mapped across {{tail_risk_quantile}} and {{temporal_aggregation_window}})
  • Operational Alert Tuning Recommendations (concrete parameter settings aligned to {{false_discovery_rate_target}})

Self-review

  1. Did the analysis rigorously verify the stability of the POT threshold selection?
  2. Is the trade-off against {{false_discovery_rate_target}} demonstrated with statistically valid multiple-testing corrections?
  3. Are all compliance references strictly aligned to {{regulatory_reporting_standard}}?
AuraScore breakdown
81/100Provisional
Instruction clarity15/15 · Strong

Explicit role, a named task, and discrete steps the model can follow.

Context architecture12/12 · Strong

Background, inputs and variables the model needs before it starts.

Constraint engineering12/12 · Strong

Hard boundaries — what the model must and must not do.

Output specification6/14 · Thin

A named, field-level shape for the response.

Reasoning structure10/10 · Strong

Ordered work items that force analysis before an answer.

Model compatibility10/10 · Strong

Length and structure that travel across frontier models.

Token efficiency5/10 · Thin

Signal density — instruction weight without padding.

Reusability7/7 · Strong

Documented variables so the scaffold adapts to new inputs.

Robustness3/5 · Adequate

Quality bar, assumptions and behaviour when inputs are thin.

Observed performance1/5 · Thin

How much real usage the template has behind it.

data-analytics
data-statistics
financial-services
extreme-value-theory
fraud-analytics
aml