Tora Adds AI Tool For TCA

Under requirements in the revised Markets in Financial Instruments Directive (Mifid II), which come into effect on January 3, 2018, buy-side firms must take a number of steps to deliver and demonstrate best execution to investors and regulators.

Therefore, investment firms need to monitor execution quality in terms of “price, costs, speed, likelihood of execution and settlement, size, nature of any other relevant consideration,” according to the rules.

Gerrit Van Wingerden, managing director of Japan at Tora says: “The key challenges of pre-trade TCA are the aggregation of all brokers and their respective algos into a common platform, the integration of the pre-trade TCA tool into a firm’s overall trading process, and the ability of the TCA tool to process massive amounts of historical data and continuously capture and integrate new order data as soon as it becomes available.”

Tora’s TCA solution is built on a convolutional neural network that is trained using real-time and historical data. Tora uses machine-learning algorithms to continuously capture new order data as it becomes available. Wingerden says the network is trained with various attributes of the order such as spread, volume and volatility.

Spread is measured by an intra-day moving average, volatility is measured by a multi-day average and volume is measured by a function of the trade quantity and a multi-day average of total volume traded. The process retrains the neural network at specified intervals by including trades executed during the current trading session in the training set.

Those factors are used to estimate the market impact of using any broker and algorithm combination, in order for traders to determine the optimal place to send their orders.

“The convolutional neural network learns by adjusting the network weights based on the deviation between the estimated average slippage and actual average slippage. Unlike most existing offerings the model is updated dynamically as new data becomes available,” he says.

Moving forward, he adds that automation and AI will play a part in keeping up with market complexity and the proliferation of information that must be processed, thanks to the rapid pace of innovation, fragmentation, and regulation.

Wingerden says the increasing number of new algorithmic trading strategies, execution venues, the number of traditional and non-traditional data sets, and the increased information requirements of new regulations have pushed firms well beyond manual and even traditional computing capabilities.

Tora is also applying “lightweight” AI to make predictions on order allocations, in order to improve data integrity and reduce post-trade issues.

The tool is available as a separate solution or fully integrated with Tora’s Mifid II solution that includes in-trade and post-trade TCA to support the execution process.