Why TCA is helping to bring a new dimension to algorithmic FX trading

May 2023 in Provider Profiles

By Professor Alessio Sancetta, Head of Algorithmic Execution Intelligence at UBS Investment Bank.

Broadly speaking what are the main goals of TCA in FX?

The goal of Transaction Cost Analysis (TCA) in FX is to analyse the cost of a trade. The analysis attempts to identify what variables affect the cost of the trade and how they do so. The analysis can go beyond the aggregate number itself, and look at individual child order execution. The goal is to provide algo users with a more in-depth understanding of algo behaviour to guide their trading choices.

Why is TCA such an increasingly important part of the FX algo trading process?

Market participants typically seek to evaluate their algo performance as well as compare different Algo providers. We believe that as FX algo volumes continue to grow, similar to what historically happened in Equities, TCA is becoming more integral to the trading lifecycle. We expect that as clients further realize the value of FX algo execution, they will demand more in-depth TCA to guide their decisions.

What sort of algo trading performance analysis is current TCA typically associated with?

TCA often focuses around the estimation of performance relative to a risk transfer price (RTP). RTP is the price paid for immediate execution. Given the decentralized nature of FX markets such a benchmark has its limits as there is not a unique RTP. Moreover, performance relative to RTP masks the actual cost of the order. When evaluating whether to use an execution algo, the user may compute the value of their alpha at a given point in time and compare this to an estimate of the cost of executing the trade. In this case, performance relative to mid-price at the time of the order (mid arrival) can be more useful. Moreover, while decentralized, there is a high level of consensus, among FX traders, that the primary market (EBS and Reuters) mid-price acts as reference price.


Competition among providers, coupled with client demand, is increasing the level of sophistication of algo execution

A great deal of effort is being made to turn TCA from purely a post-trade analytics to a pre-trade analytics tool. What impact is this having on algorithmic FX trading?

Timing, size and speed of trading are some of the parameter inputs to an execution algo, which affect the algo’s performance. The goal of pre-trade TCA is to determine how to set these inputs. For example, inputs would include user preference as clients may have hard constraints due to preferences. The choice of algorithm and its inputs is rarely based on pre-trade TCA only. As users become more comfortable with FX algo execution in FX, it is expected that pre-trade TCA will play a larger role.

What advantages does the immediate feedback of real-time TCA promise to bring to FX trading firms and especially those using algos?

Competition among providers, coupled with client demand, is increasing the level of sophistication of algo execution. Sophistication comes in the form of extracting as much information as possible from the data. Ultimately, this information is used for decision-making, which can include real-time decisions made under changing market conditions. Real-time TCA can be applied in such context. A mathematical method to solve such problems is known as Dynamic Programming (DP). DP provides that an action is taken now based on current market conditions and future expectations of the market, accounting for future actions. However, it is difficult to solve problems specifically, except for simple cases, and the size of real time information that goes through the market requires approximations. Reinforcement Learning (RL), a machine learning technique, is a way to approximately implement DP. It is likely that RL will be the methodology used in real time TCA.

There can be a lot of noise in TCA data. What can be done about that and how are leading FX providers leveraging next generation technologies like Machine Learning to help solve the issues?

FX TCA is indeed typically very noisy. This is due to the fact that time risk dominates the cost of trading. Machine learning and statistical data analysis can help extract information. However, domain knowledge is necessary to be able to curate the data and then model the cost. Purely data driven approaches would likely fail to extract relevant information. Leading FX providers would have to rely on experienced modellers and data analysts to take the area to the next level.

TCA looks like becoming more of a scientific discipline. Is it necessary for buyside firms to become experts in Quantitative data analysis in order to make the most of it?

Decisions to be optimal must be based on all relevant information. The amount of information available that can be stored in the form of data is growing exponentially. Therefore, we believe the ability to extract relevant information for a given task requires domain knowledge and data analytic skills. While buy-side firms require these skills, they do not need to acquire the domain knowledge and skills required for TCA and advanced algo execution. Instead, Algo execution providers focus on the TCA problem so that buy-side firms can dispense with becoming experts in high frequency trading and its cost analysis.

TCA is clearly not a one size fits all discipline. What factors might influence how a buyside chooses to obtain and source it?

The fact that TCA is not one-size-fits-all is particularly so for pre-trade TCA. The cost of execution in Equities is mostly driven by liquidity constraints. In fact, Equity algo providers often focus on modelling the individual child orders impact due to transitory liquidity depletion. This is less of the case in the FX world. There, there is more noise as there is more liquidity and it is difficult to pin down the cost of execution to one single main factor. In FX, an algo user might be mostly concerned with collecting passive fills without revealing information to the market and being averse selected. In short, time risk dominates the cost and makes TCA a rather challenging problem within FX.

Other nuisances typical of FX, such as venues with last look and ECN’s with multiple streams, further distance FX TCA from other asset classes. Hence, a buyside firm should realize that sourcing FX TCA from providers that traditionally focused on Equities might not be optimal.

What are your views on FX buyside firms who are looking to take a more strategic approach to TCA in order to make it a more central part of their trading systems and architectures?

It would likely benefit buy side firms to make TCA an integral part of their business. This is clearly the case when evaluating the value of their alpha. However, unless the buyside firm is already in the high frequency space such firms may potentially consider teaming up with an algo provider that is focused on innovation and research and development as a more cost effective way of using both post- and pre-trade TCA.