E.2 Reporting Line Correlation Calculations

For the reporting lines, regression coefficients are calculated using the R-model based on the threshold values. It is considered that a pairwise relationship exists between independent and dependent reporting lines.

In what-if analysis, you can make variations to the value of a variable. Variations can be applied only to the following reporting lines in the income statement:

• Interest Income
• Interest Expenses
• Transfer Pricing Charge
• Transfer Pricing Credit
• Non-Interest Income
• Operating Expenses
• Net Credit Losses
• Other Revenue

The following parameters are available in the FSI_MODEL_PARAMETERS table:

• The start date of the reference period.
• The end date of the reference period.
• The percentage of values that lie within the threshold.
• The percentage of outliers that need to be removed.

The following steps are used in the repline correlation calculation:

• Excluding Outliers: For each variable, the sigma and mean are calculated within the reference period as defined in the database. If the value of the variable lies outside the threshold provided, the respective pairs are excluded for all associated variables.

  Pairs are excluded based on the Mahalanobis distance, i.e., pairs are excluded in descending order of their absolute distance from the mean.

• Testing for Stationarity: After the outliers are excluded, the ADF test is used to check for stationarity on the time replines. The stationary is checked for each repline at two levels: I(0) and I(1). If any time repline is not found to be stationary, do a differencing of data, and repeat the test. add.test is a function of the R-library. A limitation of the R-library is that the stationary value can be calculated only if the records or data points are more than or equal to 6.

  Results are reported and used in the co-integration test.

• Testing for Cointegration: After the stationary test is done, the causal relations between regression variables are checked. Then co-integration is done.
  The following table shows the action that is performed for pairwise stationarity and pairwise integration based on the stationarity level:

  Table E-2 Testing for Co-integration

  Pairwise Stationarity	Pairwise Co-integration	Action
  Both I(1)	Exists	Do regression without any transformation.
  	Does not exist	Do regression after differencing.
  Both(0)	NA	Do regression without any transformation.
  One I(1) other I(0)	NA	Do regression after differencing I(1) series