CTR#DIV- FSMIMS-DG#1 Project was under the contract signed between Deposit Insurance of Vietnam (DIV) and FPT Information System (FPT IS). The project objective is to build a modern, centralized and integrated IT system to support DIV’s core business in accordance with the international standards.
Successful implementation of this project will provide DIV with the followings: (i) A set of software applications that support DIV to improve operational efficiency and digitalize their core business processes; (ii) A structured centralized data repository for information collection, management and exploitation (iii) An infrastructure of information technology with the disaster recovery center to guarantee data security and operation continuity of DIV.
The project consists of the implementations of 07 modules. Each and every module is developed based on the latest available advanced technologies, a combination of the international COTS (Commercial off the Shelf) solutions and the software solutions developed by FIS. The following table describes the modules and the corresponding technology platforms used.
|1||Information Management (IM)||Oracle ODI/OBIEE/EDQ.|
|2||Resolution (RL)||FPT.SmartCore (A solution platform developed by FIS)|
|3||Finance Module (FM)||FPT.SmartCore|
|4||Risk Monitoring (RM)||FPT.SmartCore|
Oracle Advanced Analytics
|6||Human Resource (HR)||FPT.iHRP. (FIS solution)|
|7||Document Management (DCM)||Newgen OmniDoc and OmniScan|
Risk Monitoring – the core business of DIV
Risk Monitoring is considered as a key module of this integrated solution. It provides significant support for DIV to enhance assessment capacity and forecast financial risks of insured institutions, thus effectively supports in protecting the interests of depositors.
Being a member of the country’s financial control system, DIV has its missions to protect the depositors, and create a common trust of the financial market in the banking sector. Therefore, DIV needs to be equipped with the system to monitor the financial risk of their customers, i.e. the insured institutions. This is performed by two measures: (i) Insured Institution Rating and (ii) Probability of Default Estimation.
Insured institution Rating
Rating system will support DI institution to analyze the situation, identify potential risks and provide accurately alerts on time to the right stakeholder. Rating is used as a basis for the DI institution to manage risk and categorize insured institutions. Rating system allows DIV to rate an insured institution using the following two approaches:
- Based on Heuristic model. This model is constructed solely based on expert’s opinions. Basically, insured institutions are rated based on financial indicators (obtained from the financial report issued, etc); non-financial indicators (organization structure, management board expertise, etc); and other information (market updates, reports issued by external sources, etc). Each indicator is analyzed, weighted and standardized in rating scale based on expert opinion to assess and evaluate insured institutions.
- Based on the result of applying mathematical models: based on the result of default probability from running regression model (logit, probit model); or the result of calculating point for insured institutions via Discriminant analysis or Factor analysis, DIV user can then translate this result into an assigned grade for each insured institution. Depending on DIV’s policy in effect at the time, translated value may vary with each grade.
Probability of Default Estimation
Based on the available historical data of insured institution collected by deposit insurance institution over the past 10 years, it it necessary to develop a model to predict PD of insured institutions in the future. The historical data is presented as a n×m matrix, where n is the number of the insured institutions and m is the number of financial measures collected for each institution. We also add an extra special column labeled as label data with the value of either (0 – defaulted) or (1 – not defaulted).
The project employs different data mining techniques such as regression and classification methods to propose to DIV an appropriate model for the time being. Selecting an appropriate model is determined based on analyzing different properties of training data set such as the amount of data, data distribution labels, data values, desired accuracy. Empirical modeling process is performed regularly to find the most appropriate model with regards to the historical data available at the time.
The process of developing a model is illustrated as follow:
After collecting, data will be pre-processed by some methods such as: standardizing data, converting data or discretizing data. The next step is to split data into 2 random subsets with a defined proportion (the commonly used proportion is 70-30). The first subset is used for training model and the latter is used for testing the trained model. Training data will be used for various models such as regression models and classification models. After the training stage completed, the model will be assigned to a specific set of model parameters. The testing stages (Example: R-Square, Mean absolute error) are used to evaluate the accuracy of the trained model. All the functions: Data pre-processing, Module training, Model testing and Model applying are implemented in OAA (Oracle Advanced Analytics).
Pham Thi Quynh – FPT IS
(Published on the FPT Technology Magazine, FPT TechInsight No.1)