How long has MN been utilizing FX execution algos and what direct benefits are you looking to obtain by using them?
Our FX execution algo platform is up and running for nearly one year and it already delivers what we were aiming for. The most important thing is that we have been able to significantly lower our costs. That is because we can change and optimize faster and have cut out the middle men, meaning we pay less fees to banks.
In addition our algo platform gives us better insight in who we are trading with and opens up a new chest of valuable data. All of our execution decisions are now backed up by larger amounts of data than before thanks to our algorithms and platform.
A nice side effect worth mentioning is that data science and algorithms are popular topics among young talent, so we can offer them exciting internships and starting positions. We are giving them an opportunity to work on next-generation execution, which is overwhelmingly data-driven in our view.
Why did the MN Treasury team decide to develop and build their own execution algos in house?
The goals that we just mentioned – lowering costs, faster work flow and more transparency – can only be achieved if you’re not solely reliant on another party’s algorithms or services. You need to have full control and ownership over what you’re working with. Therefore we decided to build the algo platform ourselves. Although it’s not always cheaper to build things from scratch, in this case it was. The reduction in fees has yielded significant financial benefits.
What made this project special for you?
We are grateful that our management gave us the opportunity to work on this innovative project and gave us the trust. You may expect such projects to be done in Silicon Valley or at international banks, but probably not so much at a pension asset manager. Achieving the Proof-of-Concept in our innovation lab with a relatively small budget and on top of our normal trading work, gave it sort of like a startup feeling. Regular pizza dinners in the lab and no guarantees the project would ever succeed. How cool is that?
What sort of algos (passive/aggressive etc) have you designed to meet your trading objectives and what key functionalities do they have?
We are in the market to execute FX spot orders that arise from the currency hedge overlay of our clients or from rebalancing flow within our client’s portfolios. Mitigating currency risk is the most important objective of our trading. Combined with the buy and hold nature of our client’s portfolios, this means that we mainly use interval based and floating executing strategies with a high percentage of passive fills. Going forward we are looking to optimize these strategies further and are also looking to add additional strategies depending on our shifting needs.
What were the main technical challenges in delivering this project?
The idea started within the Treasury team. One of our main tasks is to execute FX, so it was quite a challenge to build the IT infrastructure. Luckily we got help from an external IT partner and our own IT specialists. It was important to translate our business needs clearly into direct requirements for code development. Another challenge was to construct a highly robust risk management framework. This is vital for our pension fund clients.
How do you test the algos to see if they are performing as required?
We use model validations from external experts and universities, back testing and a whole series of technical and functional tests. And of course, we have our internal TCA which we will keep optimizing going forward. It’s important to mention that we also use BestX, the market leader in FX TCA, for assessing the performance. Through this we assess if our algos are performing as required and how they are keeping up against algos from our external liquidity providers. We want to hold our own algos to the highest benchmarking standards.
Although there was initial disbelief in the industry, based on the TCA from BestX so far, we have been able to achieve better results on a net cost basis than most of our external liquidity providers of which we are of course very proud.
In what ways are you fine tuning the algos and utilizing TCA to help achieve this?
By continuously using a feedback loop with TCA into the algo parameters. Data science has become a very important part of this process. We have hired two quant developers at our desk to do this. We believe that our updated back testing engine which allows us to run even more detailed analyses will be beneficial to the performance of our algos. In addition we also cooperate with several leading universities, such as the Technical University in Delft, and do joint data analysis projects with our pension asset management peer PGGM within our program where we offer opportunities to students in the quantitative fields to write their graduation theses. Those students also contribute to our research and the further development of our execution algorithms.
What other issues, for example compliance requirements, did you need to consider in getting this project completed?
As MN operates in a highly regulated environment, it was important to take risk management and compliance very serious from the start. Privacy, cybersecurity and market regulations were several of the many requirements we needed to take into account. Firms often see compliance as a last hurdle that can make or break a project, but we approached it very differently. We engaged our risk and compliance department from the beginning and saw them as partners, rather than ‘police officers’. We learned a lot of them and they gave really valuable advice on how the organize our project. Another thing was that there were no standard procedures in place as we did such a project for the first time. We had to figure it all out ourselves which is obviously challenging. You need to have patience and perseverance when you are pioneering.
How happy have you been with the results of this initiative so far?
Very happy! It makes our job a lot more fun because the really short lines to act, change and optimize by having our own tech at the desk. We enjoy it every day. What’s also very fulfilling is the response from colleagues, clients and the market. Our CEO asked Rick on stage during a townhall meeting last month in front of all of our colleagues to praise the project. We have also been featured regularly in the media, like on your platform and on Bloomberg. This has been quite exciting for us as our firm generally tends to operate more on the background. Although it may sound a bit boring, we are also very satisfied that every order that has been executed in the market did not cause any issues for MN or our clients, who were supportive from the beginning and are also very satisfied with the results.
Do you have plans to build upon the work you have done so far and will the team be looking to roll out more FX algorithmic toolsets?
We are always looking to optimize our solution and assess whether we can leverage our existing technology towards other instruments, asset classes or processes within our organization. We’ve learned a lot from handling the big data that’s acquired through our platform, and the lessons learned from this quantitative approach can be applied to many areas of the investment portfolios and procedures of our clients.
In addition, we are currently further developing our in-house TCA tool together with the Business Analytics team of our IT department. This increases the granular insights that we can gain from our TCA and helps compare to external LP’s even better. The cool part is that this also generates recommendations on which algos to use, including external ones, by making use of a refined recommendation algorithm that uses AI techniques. With this tool we are already leveraging our tech into other instruments like FX swaps, which is a new TCA-module currently under construction.
Developing all these things in-house means that we have a shifting interest in the type of partnerships we engage in: based less on your classical banking relationships and off-the-shelf products and focused more on for example trading venues, liquidity pools, cloud, tech and data providers.
Would you recommend that other firms in your industry increase the algorithmic trading component of their FX execution policy and what advice would you give to those considering taking on the task of building their own algos?
If a firm has enough flow in FX or other asset classes it can certainly be interesting, but be aware that it takes a lot of work to build it from scratch. Our algo platform is more than just the pricing logic in our trading engine, but also the fact that we have direct market access and an entire platform to support these things, including STP connections to our order manager and venues, automatic allocation of the execution flow to our credit intermediaries, tailor-made and highly intuitive user interface that can be customized further very quickly and easily when our requirements change.
We are a not-for-profit pension asset management firm that is not looking to commercialize our solution, so we are open to sharing knowledge and technology with others within the pension fund community. The exact form in which this could take place is dependent upon many factors, such as the knowledge and technology that we can obtain from peers in return and the possible economies of scale that can be realized by co-developing and or enhancing certain aspects of the algo ecosystem further with other peers.
What we see is that everyone is currently sitting more or less on their own ‘island’ dealing with the same challenges, so to increase reaching out to each other and sharing knowledge makes a lot of sense. If you’re reading this and would like to find out more, feel free to contact us so that we can start the conversation.