"Whenever you eliminate the impossible, whatever remains, however improbable, must be the truth" – Sherlock Holmes (care of Sir Arthur Conan Doyle)

Corroborate - to strengthen or support with other information/evidence i.e. to make more certain results of analysis (facilitates Collaborative which is to cooperate or to work together)

Analysis - an investigation of the component parts of a whole and their relations in making up the whole

The essence of Corroborative Analytics (derived from Corroborate and Analysis) is to give more weight to information or evidence gleamed from multiple sources. The adoption of Corroborative Analytics is critical for firms to ensure information is C.A.V.E. compliant.

To best understand the term, Corroborative Analytics, it is advisable to look to the law (after all one use of these analytics is for regulatory reporting).

Evidence – Prima Facie?

In court some types of evidence are admissible but are "suspect". This means that they often turn out to be untrue or incorrect (compare this to self assessment/ profitability forecasts/ bad debt reserves information etc…).To protect the accused against being convicted (compare to senior management under Sabanes-Oxley) on the basis of suspect evidence, safeguards have been put in place (cf Controls). The most important safeguard against suspect evidence is the "corroboration warning" (cf alerts).

Corroborative evidence

Corroborative evidence is "additional" evidence. It is evidence in a case which:

• Implicates the accused in the crime (i.e. if all information was analyzed then the issue would have been identified)
• Is admissible and credible (Re McCarthy (1970) - Corroboration is of no avail if the claimant's story is not believed)
• Is independent (i.e. does not necessarily require another witness who swears to the same thing… Fair inferences of fact arising from other facts proved, that render it improbable that the fact sworn to be not true, and reasonably tend to give certainty to the contention which it supports, and are consistent with the truth of the fact deposed to, are, in law, corroborative evidence." (Ritchie, C.J., in the case of in Re Estate of A.K. MacDonald (1922-1924))

An example of Corroborate Analytics inputs for Operational Risk

The above diagram illustrates how corroborative analytics is used in the validation of Operational Information pertaining to Risk.

In this instance the following information is taken from the above information sources relating to retail branch banking

• Self-Assessment – Branch managers are asked about the controls over life assurance policy sales in their branches.
• Scorecard – Branch managers explicitly asked to rate controls over policy sales at their branch
• Manual "loss" – Key Risk Indicators (KRI) designed to highlight controls over policy sales are developed (average policy life, number of policies per month that are declined in the 14 day cool off period, "churning" – conversion of one policy to a similar one etc…). Managers are asked to submit this information on a monthly basis via an automated web based capture tool
• Automated loss – The KRIs described above are captured automatically from underlying systems (such as policy databases)
• Other sources – These include General Ledger (costs associated with policy closure), Commission Systems (corroborate commission paid versus number of policies in database), Human Resource databases etc…

Once the information above has been gathered various mechanisms are used to corroborate information as follows;

Self-policing (internal corroboration)

Self-assessments

• Comparing list of controls for each branch for each activity (policy sales) highlights control information omissions (or submission)
• Comparing controls and their associate stated control ranking(Hi-Med-Lo or Ranked) used to illustrate reported deviation from norm

Scorecards

• Comparing controls and their associate stated control ranking(Hi-Med-Lo or Ranked) used to illustrate reported deviation from norm

Manual "Loss"

• Comparing Key Risk Indicator data per branch submitted used to illustrate reported deviation from norm

Automated "Loss"

• Comparing Key Risk Indicator data per branch calculated using rules engines used to illustrate reported deviation from norm

Whilst internal corroboration is a useful tool it, as in the legal profession, is superceded by the need for vilification through the use of "external corroboration"

Policing ("external" corroboration)

Self-assessments

• Comparison of self-assessment of controls to associated KRIs e.g.

By linking the self-assessments to associated KRIs it appears that there internal discrepancies in this 2 branch case study. Though Branch A rated themselves highly on their controls the number of cancellations is higher than in Branch B (which rated themselves lower on controls). Does this point to a problem with the rating system, the manager or some other factor? As the number of policies lapsing in the cool off period is significantly higher in Branch A than Branch B corroborative analytics may appear to point to control weaknesses or even unethical behavior in Branch A! How do we investigate further?

Manual vs. Automated Loss Submission

• Comparison of manual versus automated submission of KMIs

One of the elements that need to be validated is the accuracy of the rules engine in scenario generation. If the underlying assumption or inputs in the automated engine is flawed then information generated by it will not be C.A.V.E. compliant. In the above it appears that the rules/ scenarios/ inputs are accurate as they tally within acceptable parameters for Branch B. However there appears to be discrepancies in manual vs automated reporting for Branch A which could point to fraudulent reporting.

Scalars

Even though, in absolute figures, it appears that there are issues with Branch A there may be logical reasons for the discrepancy. To determine if this could be the case, scalars can be added to "weight" these issues

When scalars are brought in it appears that;

• Branch A is significantly larger in size than Branch B with nearly twice as many staff and triple the number of active policies – this could explain the higher number of lapses.

On bringing in metrics related to branch size

The prorated figure is now more in line with expectations but still high when compared to the self assessment control score. However this may indicate managerial inexperience rather than mis-management

• Branch A has less experienced staff – this could explain both the higher number of lapses and the discrepancy between manually reported and automated figures. When coupled with the information above regarding discrepancy in self-assessment scores this would corroborate the theory regarding managerial experience (this may further be corroborated by incorporating training taken information).

The above illustrates how Crest Rider can develop infrastructure, business rules and intelligent management analytics to both corroborate information and provide real business insight into the state of an organization.