×
Home Companies Industry Politics Money Opinion LoungeMultimedia Science Education Sports TechnologyConsumerSpecialsMint on Sunday
×

Greater computing power needed

Greater computing power needed
Comment E-mail Print Share
First Published: Wed, Feb 09 2011. 08 36 PM IST
Updated: Wed, Feb 09 2011. 08 36 PM IST
Risk modelling is now employing advanced methods such as the Monte Carlo simulation and neural networks that require significant investments in servers, data storage and computational power. As commonly understood, the neural network method is a non-parametric process and it is employed when the data cannot be identified with a recognized distribution. Credit default is one such distribution that occurs infrequently at distant intervals.
Therefore, while modelling probability of credit default, statisticians tend to use non-parametric methods such as decision tree, support vector machines (SVMs) or neural networks. Since a neural network is a black box unable to relate the result with logic, it is used as a last resort. Incidentally, these are the suggested statistical approaches mentioned in Basel-related papers. Leading banks in the US and Europe use SVM learning for calculating probability of default and loss given default.
Grid computing, another trend that is catching on, is a combination of multiple computer resources from various administrative domains to reach a common goal. It has gained a lot of significance with real-time analysis of large volumes of data. For instance, in the process of calculating portfolio VaR (value at risk) and potential exposure using Monte Carlo, which require large volumes of data processing and simulations, grid computing is widely used. As banks move towards advanced risk-measurement approaches, advantages of grid computing will be fully realized.
The Monte Carlo method uses random numbers and probability to solve problems. It is often used when the model is complex, or involves more than a couple of uncertain parameters. A simulation can typically involve over 100,000 iterations of the model. By using random inputs, we essentially turn a deterministic model into a probabilistic one.
In order to perform enhanced risk modelling needed today, banks will need far greater computing power—approximately 5x to 10x what was used in the pre-crisis era. Leading global investment banks use grid computing for complex trading-related risk analysis spread across multiple geographies.
Indian public sector banks are only now making the transition from regulatory compliance to developing sophisticated risk infrastructure. The new private sector banks have taken a lead in installing state-of-the-art infrastructure for banking and risk management. Banks in India are, therefore, yet to move to grid computing, though Western Europe and the US tier I banks have been very advanced in leveraging those methods.
N.G. Subramaniam is president of TCS Financial Solutions, a division of Tata Consultancy Services Ltd, that has implemented several end-to-end solutions for banks.
Comment E-mail Print Share
First Published: Wed, Feb 09 2011. 08 36 PM IST