Kasper please tell us a little about your day-to-day responsibilities within Nordea Asset Management.
Based in Copenhagen, I am heading the FX trading desk, a cross asset quant team, and acting as product owner for the technology organisation being responsible for digitalising our investment process. My different roles reflect how we as a modern buy side trading organisation combine and utilize market knowledge, quantitative frameworks, and technology to achieve the best possible outcome for our clients.
Broadly speaking what are the key objectives of your currency trading team?
The FX trading team’s fundamental goal is to handle the FX business related to Nordea Asset Management’s investment funds as efficient and cheap as possible. We are involved in the whole investment process, from an order’s initial creation to its trading. The trading is a subset of what we do, and we try to optimize the whole process, not just the trading part. In relation to trading, the goal is to execute orders efficiently to the lowest cost possible with respect to the order’s underlying investment objective and reduce operational risk. Additionally, we are also responsible for the majority of the day-to-day cash management and hedge adjustments for our investment funds, although this is by and large handled automatically by an internally developed algorithm. The FX traders are the experts in all things related to FX, so their role is also to advise portfolio managers on liquidity conditions and the optimal execution of trades as well as hedge adjustments.
Unlike many buyside traders you have a great deal of experience of sell-side operations having been head of e-FX at a major bank. Having seen both sides of the fence how has that helped you with your current job?
With more than 10 years’ experience on the sell side working within FI and FX e-Trading I have seen the impact of how technology, structured trading, and quantitative frameworks changed our Markets operation, but also how this transformation helped many of our clients. I believe this knowledge and many of the best practices we applied in the trading teams can be adopted by the buy side to a large extent, resulting in a higher quality of execution, lower transaction costs, and reduced operational risk. Particularly we have changed our way of working between traders, dedicated trading quants and technologists, working as one team, physically seated next to each other. A more focused way to combine these skillsets have enabled us to automate large parts of our order generation and execution, freeing up time for traders to focus on the harder to execute trades.
How enthusiastic has Nordea Asset Management been towards the use of FX algorithms and in what ways are they a good fit for the type and style of trading your team are carrying out?
We have used FX algos for several years now, and as Nordea Asset Management’s AUM has grown significantly over the past years, FX algos have had an increasingly important role in the execution of larger orders. As most of our FX orders have a passive execution objective, FX algos are well suited for that type of trades. We have been working hard over the last two years on building a quantitative framework for using FX algos. This includes the implementation of an algo wheel, doing TCA, monitoring and reviewing performance periodically, and discussing it in structured broker reviews with our counterparties.
What types of FX algorithms are you currently employing and what factors generally influence this?
The style selection depends on the investment objective of the order. We generally use either passive (and “opportunistic” passive) or implementation shortfall algos, with the vast majority of our trades being in the passive category. This gives us the possibility to execute a trade over a longer time horizon in order to minimize market impact and increase spread capture. Of course, there is the well-known trade-off between spread saving and market risk, which we try to find an optimal balance in. This optimum is a moving target depending on the current market conditions and spreads. Another deciding factor for the type of algorithm used is the time of the day, along with the idiosyncratic factors of the currency pair in question, as well as the time until Europe winds down business. We don’t leave algos executing overnight without oversight, so we aim to be done before the London session ends where liquidity tends to dry up for many of the currencies we trade.
To what extent does an understanding of market dynamics and the liquidity landscape influence the decisions you take on how to deploy an FX algorithm and the execution style that may be most appropriate?
We use pre-trade analysis to get a grasp of the typical liquidity conditions and intraday volume profile for a given pair. Our trading is concentrated in the most active hours of the day, so it’s only the more exotic currencies that we tend to spend energy on pre-trade to make sure we execute in a liquid time window. However, we still believe that an algo should be intelligent enough to adapt itself to different liquidity conditions and act accordingly. So, if there is less than expected volume in a given day, the algo should adapt to it and therefore we will normally not change our execution process due to varying liquidity conditions. Still, there are some currencies (for example pegged currencies such as the DKK) where algos are not very applicable, and hence we handle these in a different manner than the rest.
Many asset managers use algos provided by banks. How does Nordea Asset Management source yours and what are the reasons for that choice?
We believe that the banks are and always will be the experts at developing execution algos. They have more data and resources allocated to developing algos than we will ever have, and so we don’t believe we can compete with them on that front. We have therefore decided that our resources are better spent elsewhere. So, we use the algos provided by our top counterparties. At the same time, it’s important that we clearly communicate our investment objectives to the banks and that we together map those to their specific algos in order to select the best-suited ones.
As previously mentioned, we have implemented an algo wheel for our passive and implementation shortfall trades. What this essentially does is that it assigns a probability to each broker (which can vary depending on their performance) and allocate orders to them based on that. Some algo users might critique this approach and say that they would never make a ‘random’ choice in selecting an algo. The counter to that is, that we have introduced structured randomness as a conscious choice to avoid trader bias – we control the distribution of algo trades based on historical data and not by individual discretion.
What are you particularly looking for in terms of the functionality that FX algos can offer?
The ability to access many different liquidity venues as well as the capacity to internalize flow are two of the key benefits of using an algo versus a risk transfer. We also have several hygiene factors related to functionality that an algo provider must be able to deliver in order for us to use them. These include the ability to upload trade data automatically to our TCA providers, automated forward roll after the algo completes the spot execution, the ability to trade synthetic crosses, and last but not least be able to share knowledge around executions and market trends, etc.
How involved do you get in the testing and fine-tuning of the FX algos you are using?
We provide extensive and structured data-driven feedback to our algo providers so they can guide us in using the optimal parametrization for their algos. We believe they know their algos the best and therefore are in a better position than us to determine exactly what is needed to optimize performance relative to our execution objectives.
What are the key benefits that you are getting from algorithmic FX trading?
The main benefits of algos are reduced market impact and lower transaction costs on larger trades that we would otherwise have to manually split up and work over the day or trade as full amount risk transfers. The algos we use have access to many different liquidity venues, and a good part of the volume is internalized which is great as it minimizes market impact.
How “hands-off” are you prepared to be once you have committed to using an algo or do you still like to micro-manage various parts of the execution process?
We are completely hands-off during the algo execution. If you micro-manage and intervene in its execution it is impossible to preserve consistency in the trading data and as a consequence you end up compromising the possibility of learning from the data because of the bias embedded in it. What probably distinguishes algo users on this topic is that some try to optimize the outcome of every trade and will thus tend to intervene and micromanage the trades. Whereas on the other hand, we want to have a systematic and quantitative process that enables us to learn from the data and improve our process over time. And we have already seen improved results with this approach in terms of lower transaction costs and reduced market impact over the last year. But it is an iterative process where we constantly learn by experimenting and that sometimes includes doing things that might seem less optimal in a specific situation.
Many FX algorithms now perform very similar functions. How do you go about establishing and focusing on the ones that will be most suitable for your own requirements?
It is difficult to statistically rank algos against each other because you need a very large sample set of trades in the same time period and market regime. Because of that, we mainly use a via-negative approach to find out if certain algos have unwanted behavior or are performing significantly worse relative to the peer universe. The challenge in doing that exercise is the lack of data. As XTX has recently shown in a study, very few asset managers will have sufficient data samples to statistically determine whether an algo’s performance is good or bad because of the large variance in the data. The best way to address this challenge is to use aggregated peer data. Third-party TCA providers such as BestX and Tradefeedr aggregate data on behalf of their clients which makes it possible to compare performance across providers. However, the problem is that today we cannot be certain that we are comparing apples to apples, since there is still some lack in the standardization of FX algo data (one needs to control for the use of different urgency settings, intra-trade changes, and various other human interventions). But hopefully, that will be solved in the near future.
Some firms rely on their own internal TCA toolsets whilst others work with external independent third-party TCA providers. How do you approach this task?
We only use independent third-party TCA providers, as it provides us with a standardized data model, access to peer data, and pre-trade analytics. We use BestX, but have also started using Tradefeedr who has some neat features. In addition, we also do further analysis ourselves in Python if we want to conditionalize the data in a certain way or integrate other proprietary variables.
How much has the Covid-19 pandemic and working from home environment acted as catalyst for your team to increase their level of trading automation including the use of algos?
When fears over the spread of Covid began to really impact markets and liquidity in March 2020, we saw spreads widen significantly along with high volatility. This caused us to increase our algo usage and we saw that algos did a very good job of outperforming the risk transfer prices, even by a wide margin.
The Covid event also prompted us to initiate several transformative processes at the trading desk for how we work, effectively shifting our mindset towards a more systematic way of trading compared to a discretionary one. In doing so, we have automated many of our previously manual processes related to order generation and implementation, which have freed up a lot of time that we can now spend on other value-adding tasks, as well as reduced operational risks associated with manual errors. It also included bringing in new talent with the capabilities that are needed on a modern trading desk such as being able to code in Python, having strong quantitative abilities as well as a traditional trader skillset.
What made this transformation possible is a strong strategic focus from Nordea Asset Management who invested resources into improving the end-to-end investment process. This involved building up a 15-man dedicated technology organization that sits close to the trading desks.
They help us develop and streamline our investment processes and are the key success factor for enabling our trading automation. We also have a strong cross-asset quant team who collaborate with the three trading desks (FX, Fixed Income, and Equities) on building and enhancing our structured trading frameworks, evaluating our trading behavior, and ensure that our decisions are based on data.
In what ways do you think work could be undertaken to further improve FX algorithms to make them even more powerful and flexible for firms like yours?
We think it’s more about how buyside firms can improve their processes and frameworks to better incorporate and use FX algos in a sophisticated way, as opposed to the other way around. It’s ultimately a cooperation between the buyside firms and the banks, and the buyside have to be able to clearly state and communicate their goals and execution objectives to the banks, as well as give them structured feedback so they can adapt and advise accordingly.
As a leading European asset manager we have the necessary in-house capabilities that are needed to build a solid foundation for sophisticated algo trading, as well as being able to experiment with new solutions and initiatives. For smaller asset managers who might not be able to do the same, enhancing the use of data sharing amongst buyside firms could lower their search cost in terms of specific algo selection without having to go through the initial process of experimenting.
Two other points to improve are 1) a greater level of data standardization and sharing between providers and users, and 2) the challenge of making the forward roll of the algo more competitive, as it is generally only traded with the algo provider after the spot execution, and therefore not in competition. That can potentially be quite a drag on performance in forward algos, as it doesn’t really matter if the algo saves a few bps of spread on the spot leg if you then get a poor price on the forward points.
What steps do you think banks and algo providers can take to increase the appeal of algorithmic FX execution and to make it more accessible for a wider range of buyside firms?
Standardization, transparency, and proper disclosures from the banks are essential elements to ensure a good and trustworthy relationship. Furthermore, it’s important that the provider map their algos to different execution goals and ensure that the client has realistic expectations about performance. We also believe it’s important that the banks are proactive when it comes to providing advice pre and post execution, and that of course requires that the bank understands the client’s execution goal. Finally, data sharing among buy sides would help lowering the entry barrier.
Looking to the future in what ways do you expect Nordea Asset Management to be increasing its use of FX algorithms?
As noted previously, we are already using FX algos extensively. One area which is in rapid development is the use of NDF algos. That space is still in its early stages, but we expect to make use of them in the near future for some of our larger tickets.