Chapter 19 - Data Flashcards
Personal data
Allow an individual to be identified
Many protection laws
Sensitive data is more private
Big data
Large data sets, different sources, analyzed quickly
May or may not include personal data, use anonimisation
Be transparent when collecting
Data governance
Management of availability, usability, integrity and security of data
Should have robust data governance policy otherwise, legal, operational, reputation al and expense risks
Data governance is key during merger and acquisition
Data risks
Inaccurate or incomplete Not credible Not sufficiently relevant Past data does not reflect what will happen in the future Chosen data groups might not be optimal Data not available for intended purpose
Algorithmic decision making
Automated asset trading is example where model risk arises
Data requirements
Sources of data
Publicly available
Internal
Industry wide
Industry wide data collection schemes
Not precisely comparable
Companies operate in different geographical or socio economic sections
Policies sold by different companies are not identical
Sales methods are not identical
Companies will have different practices
Nature of data stored will not be the same
Coding for risk factors may differ
Industry wide data collection schemes
Additional problems
Data may be less detailed/flexible
Data may be more out of date
Data quality may be poor
Not all organisations contribute
Checks on data
Recon of member/policy numbers
Recon of benefits and premiums
Movement data against accounts
Validity of dates
Consistency of contribution
Consistency between sum assured and premium
Consisteny of asset income data and accounts
The recon of beneficial owner and custodian records where assets owned by third party
Full deed audit for certain assets
Consistency between start and end period shareholdings
Records picked at random for spot checks
Lack of ideal data
Insufficient volume
Data has not been captured at a sufficiently detailed level
Risk classification and reduction of heterogeneity
Aim to have homogeneous data since heterogeneity distorts results
Removal of heterogeneity needs to be balanced against having sufficient data in each group