This is a demonstrative model for calculation of Expected Credit Loss (ECL) of Agriculture Sector of Pakistan. The model is only for demonstration and may only be used for educational purposes. Due to lack of information, a lot of information used is estimated or calculated.
Expected Credit Loss (ECL) is the calculation of amount expected to be lost on a loan in present value. The formula for ECL is ECL = PD x EAD x LGD x DF
PD means probability of default of a party involved in credit, the borrower.
EAD means the exposure at default, the amount outstanding on loan at the time of default. EAD is the present fair value of the asset known as Replacement Cost (RD) plus any future increase in the value of that asset also called Potential Future Exposure (PFE).
LGD or Loss Given Default is the expected percentage of loss on default.
DF or Discount Factor is the value used to calculate present value future default.
Variables have been identified through statistical analysis and various studies. After identification of the variables, they are input into the model which is then trained to predict the loss. The model is based on Machine Learning, a single layer Artificial Neural Network (ANN). It is trained on historical data and validated to perform prediction on future data as provided by the user. Among other advantages of using ML is that ML based model is less complex and involves almost no calculation at user's end for example, you don't have to arive at expected loss by first calculating other required variables such as PD, You can directly predict the amount and percentage of expected loss.
In ECLM we take into consideration all the variables that affect the Expected Loss. In addition as a forward looking model ECLM is required to be calculated on the basis of historical and future data. There are various research studies and information provided by the relevent departments that help us identify and narrow the variables that must be included in the model. For example following list of variables was published by State Bank of Pakistan in one of their studies.
List of variables used in this research are described in charts below.
Expected Loss is not the same in the whole country. It varies region to region based on many different factors. With the use of past data we can calculate and predict the probability of default and expected loss in a specific area where loan is being disbursed. We arrived at following region wise calculations of loss by analysing agri credit data starting from as far as 2010.
|Geographic Agri Credit (Expected Loss)|
|GDP Growth (Agriculture as % of GDP)|