28 Loss Forecasting

Apart from calculating the provision, by EL and IL approach, OFS Loan Loss Forecasting and Provisioning forecasts the losses by using ratings or days-past-due matrices based on the number of customers or the total amount of exposures across product types. The loss forecast component doesn't report the losses for the future period; instead, it predicts the status of the exposure count or exposure amount.

For example: For the current period if the total exposure count at a given product type is 3000 and the forecasted PD for period 1 is 10%, then the loss forecasted value would be 300, then 2700.

The user input matrices would differ in their frequency ranging from a month to one year. The forecasted period is based on the least available frequency to five (5) periods for rating based and twenty-four (24) periods for days-past-due (DPD) based. The loss forecasting procedure is computed as follows:

• Determination of Min Frequency: Minimum frequency period of the matrices for both rating-based and days-past-due-based is used as an input for the Poisson process, to bring down all the other matrices to the common platform of frequency.

  For example: For a given set of exposures if the matrix frequency period ranges from Monthly, Quarterly, Half-yearly to Yearly, the minimum frequency period of all the matrices available (monthly) will be used as a base frequency for the other matrices to undergo the Poisson process. The forecasting is done for five (5) months for rating based and twenty-four (24) months for DPD based, excluding the current period.

• Loss forecast for Current Period: For current period values, the LLFP application will just populate the summation of the values on the given dimension. This will not need any matrix intervention. Normally, the loss forecast is done on predetermined dimensions like product type, product, asset class, and so on. Hence, while reporting the current period; LLFP will sum up the values across the selected dimensions for both exposure count and exposure amount level.
• Poisson Parameter: The Poisson process is initiated after the successful assignment of Individual exposures undergoing the Expected Loss or Incurred Loss approach to the transition matrix. The matrix is assigned based on predetermined dimensions, Customer type, product type, and currency. All the matrices irrespective of the frequency applicability will undergo the Poisson parameter.

  Poisson Parameter = 1-exp (-?) = ?; where ? = the probability of default values for a given period.

• Calculation of Probability of Defaults: The default values for the forecasted period, 5 periods or 24 periods, are loaded by using time-homogeneous and time-non-homogeneous matrices. For those matrices with variant frequency, the Poisson process of decomposition is used to trickle down the matrices to a common platform of frequency and then loaded for the respective periods.
• Customer count & Exposure amount: The LLFP application supports the forecast based on the customer count and exposure amount. Under the given dimensions, Product Type, Geography, and so on, the sum of exposures or amount of exposures is multiplied with the corresponding default values. For the second consecutive period, the output of the first period is multiplied by the corresponding default values, and so on.