150 likes | 290 Views
New validation method and IT system in DK. Søren Burman Nordic meeting 2014, Tórshavn. Outline. The validation system Where to make the cut… Manage and edit cases Future plans. The validation system. The automatic validation system consists of three procedures.
E N D
New validationmethod and IT system in DK Søren Burman Nordic meeting 2014, Tórshavn
Outline • The validation system • Where to make the cut… • Manage and editcases • Future plans
The validation system • The automaticvalidation system consistsof three procedures. • For eachenterprise, the data is testedagainstitshistoricalreportson the service item level • A score is applied to all observations basedon: • How large is the given observation compared to earliervalues for the given cell. • How large is the normal variation for the service item in question. • How large is the observation in absoluteterms.
The validation system • For each enterprise, the data is tested against its historical reports on the total flow • This is done in order to capture missing trade • Similar scoring system as in procedure 1, but with a lower ”impact” • Captures enterprises that suddenly report zero trade
The validation system • Check for use of new service items • If a new service item code is used that has never been used before, it cannot be tested against its historical reports • Captures errors where enterprises use a wrong code, but maintain the same level of trade • New reporters willnot betreated in the validationsystem untilsixreportsareapproved • All new reporters must bemanuallyapproveduntilsixapprovedreportsareavailable
The validation system Manual approval list Non validatedreports Lessthansix approvedreports Automatic validation system Cut off What is considered ”wrong” enough Possibleerror list – sorted by value Procedure 1 Procedure 2 Procedure 3 Possibleerror list – sorted by score
Where to make the cut… • Not all deviations are errors • Score gives an indication of severity • Below an estimated view of the hit-rate for the year 2012
Where to make the cut… • A case is defined as a report from an enterprise • Many of the suspicious lines are linked to the same enterprise, so the actual number of cases is lower • How many cases do we “usually” process Cut of from above, knowing that many suspicious lines are not errors
Where to make the cut… • All cases having at least one line with a score higher than 15 are handled by the case reviewers • 30 to 40 cases pr. month ~ 10 pct. of monthly reports • On top of the that: ~10 cases from procedure 3 • Ad hoc chosen levels: • Above 10.000.000 DKK OR • Above 2.500.000 and more than half of the trade • Roughly put: 50 pct. of the cases are hits.
Manage and editcases • All cases are stored in one system • Grouped by enterprise • Sorted by highest score across the reports • Different filtering options available • Automatically generated cases >< Manual cases • Suspicious lines >< New service codes • New reporters • Blocked or non-blocked cases • Case worker identity
Manage and edit cases • Double clicking on a case brings up the overview of the enterprise • Errorsare marked with red • Yellow indicatescellsthatare changed • Purpleindicatesthat the reportsareblocked
Manage and edit cases • Edit data by double clicking on a cell • Enterprise and period • Data for given cell • Development over time • History for the given cell
Future plans • Simplify and enhance the filtering options in the overview of the cases • Add more automatic validation procedures • Merchanting • The geographical level (especially country codes 4A and 7Z) • Construction • Insurance
Validating trade in services • How is ITS data validated in your country?