A Tier 1 South African bank needed to validate the market data time series used in their Value at Risk (VaR) and Capital Calculation processes. They also needed to establish control methodologies around the market data time series.

Solution:
Leveraging data science techniques – long short-term memory (LSTM) neural networks, autoencoders and convolution neural networks – the Riskworx team developed tools to flag inconsistencies in the client’s data sets in support of the validation process. The tools also suggested data fixes and alternative market data, as well as predicting VaR and Profit & Loss (P&L).

“The tools also suggested data fixes and alternative market data, as well as predicting VaR and Profit & Loss (P&L).”

Value
The client was able to remediate their data sets, ensuring more accurate VaR & Capital Calculation results. This meant that the client was able to save a considerable amount on their Risk Weighted Assets (RWA). In addition, the client’s traders were able to more accurately hedge their positions based on the predicted VaR.