ING’s AI is smarter at pricing bonds


By Flora McFarlane.

ING has launched Katana, a new artificially intelligent (AI) bond trading tool which uses predictive analytics to help price trades for clients. The bank claims that its new web-based ‘AI trading assistant’ allows its sales-traders to assimilate relevant trade data more efficiently, leading to faster pricing and ultimately more liquidity.

The interface provides an overall view of what a trader might need to know for the trade, including current position, impact of request and the relevant price history.

Santiago Braje, global head of credit trading, at ING said, “Using historical information and real-time information, it brings forward-looking predictions. The algorithms make predictions about what the winning price will be, which the user can utilise to price a trade.”

The platform was tested in controlled conditions at specific points in the day, measuring speed, accuracy and consistency of pricing. The results showed that Katana provides faster pricing decisions for 90% of trades and contributes to a reduction in trading cost of 25%, and was four times more frequently able to offer the best price to clients.

Leveraging technology is vital to help traders join the dots in traditionally over the counter (OTC) fixed income markets. Compared to the common equities markets, fixed income is significantly less liquid with far less frequent activity in any given security. As a result a wide data set needs to be considered in any pricing decision. Initially ING chose the Emerging Markets desk in London to test out Katana.

Braje said, “In a way, we set ourselves the more difficult challenge because it is one of the more illiquid markets, and if we find it’s effective in illiquid markets, it’s a good indication that it will be effective in liquid markets.”

The London-based bank also reported that it is constantly learning which features are the most relevant and looks for factors that can influence pricing. The tool draws on a wide range of datasets, including five years of historical trades, which provides examples on trades won and lost.

Frank Derks, head of advanced analytics and artificial intelligence at ING Wholesale Banking said, “We try to feed as much data in as possible, find out which features are the most defining ones. We have been surprised at which data source was the most relevant. We don’t try to predict; we try to understand what the machine comes up with and integrate the algorithms further.”

ING stressed that the technology is an augmentation of current human capabilities. Users are able to follow through on or deviate from the information the tool provides, depending on what information they have at their disposal.

“We are combining AI with human intentions,” said Braje. “If we look at the algorithms by themselves, there is space for improvement, but they cannot match the humans in terms of accuracy. However we do see that the human with AI performs better than the human without.”

Roll out across other fixed income asset classes is expected in 2018, with an end game of delivering pricing to buy-side clients. While Katana enhances the bank’s ability to provide competitive trades, the tool will also tailor the quotes depending upon the client, as the client is a factor assessed by the trading algorithm, so buy-side firms should not assume they will all see the same price for a trade.