RISK MANAGEMENT ANALYTICS
Many companies "talk" about risk management, create lists, define mitigation strategies and make use of consultants to support them. But only fewer companies use the available data to quantify risks (financial and non-financial) and to support decision-making processes in a proactive and modern way.
Small and Big Data Analytics tools can be used to:
- Quantify risks based on historical data, information from similar organizations or third-party data
- Use information on historical risk patterns to predict future risks
- Build models to support investment decisions, taking into account costs and risk probabilities
- Develop risk models that support managers with limited historical data
- Use behavioral models to investigate emerging patterns, identify anomalies and anticipate future modes of risk.
The primary objective of an analytical Fraud Detection project is to summarise, through a series of specific scores and alerts, the key variables that can anticipate fraudulent events. The analysis goal is to provide the Management with a series of tools for the preventive (and continuous) fraud management.
There are many possible fraud patterns and they change rapidly and continuously. In particular, there are two points that need to be examined with the help of statistical models:
- Anomaly detection: the identification of patterns and any anomalies using channels for banking activities is fundamental for the bank; it would be useful to model and assign a score to customers and/or accounts that presents changes or anomalies in the pattern of use of banking services.
- Mule account prediction: these are Current Accounts used to raise funds from fraudulent activities; to sum up, these accounts are used for the last step of the fund subtraction. It is therefore essential to identify the current accounts that are likely to be Mule Accounts;
The implementation of fraud forecasting models, assigning to each customer/current account the probability of being connected to fraudulent activities, is carried out through statistical processing of particular behaviors and characteristics (anomalous movements and/or presence of characteristics detected as risk factors, etc.). These behaviors of individual customers become manifest in a predetermined period of time, before the prediction moment.
Credit risk management
The development of the statistical procedure for the credit risk quantification, according to the Internal Rating Based approach, requires the construction of statistical models suitable to estimate three indicators for each exposure in the portfolio:
- PD (probability of default)
- LGD (loss given default)
- EAD (exposure at default)
For the estimation of each of the three components, one or more scoring models are designed according to the homogeneity of the portfolio exposures default rates.
In the construction of an internal rating system, the main aspects to be considered are:
- Definition of default
- Sample extraction for model development
- Analysis of the historical depth of internal behavioral data
- Quality of available data and selection of useful information
- Evaluation of data acquisition from external databases