ETL (Reference Implementation)
The ETL (Extract, Transform, Load) layer provided with the accelerator, using the default file format, will perform the following transformations when data is loaded from the input files into the datastore.
Risk Factor Name
If the risk-factor name is not included in the input file (using the default file format), a name is generated as follows:
- Delta: Tranche Name + Curve Type
- Vega: Tranche Name + Curve Type
- Curvature: Tranche Name
Normalisation
In the default file format, each row of the sensitivities' files contains the sensitivity as well as a description of the risk factor and tranche. The data normalisation splits this information across three stores in the datastore, as described in the data model above.
Vectorisation
For efficiency, Delta and Vega sensitivities are stored in vectors. The entries of the vectors represent the tenors/maturities of the risk factors.
In the default file format, we allow sensitivities to be provided as either single values or as vectors. During the ETL, sensitivities are grouped together into vectors.
Interpolation
For compatibility with risk engines, Delta and Vega sensitivities may be loaded for any tenor/maturity. During the ETL, these sensitivities are interpolated to match the tenors/maturities in the specification.