Ronald, please tell us a little about your day to day responsibilities and tasks within APG?
At APG we have a global trading desk with traders in Amsterdam, New York and Hong Kong. The guiding principle of the trading desk is to implement exposure of clients through the execution of orders at the lowest cost and with the least possible market impact. We trade equity, fixed income, money markets, commodities, FX and derivatives on many assets. For the past 7 years I have been responsible for the FX trading and execution for our pension fund clients and our mutual funds. These responsibilities range from hands-on trading to advising clients on their currency hedging mandates. It is my job to develop, analyse and choose trading methods, liquidity providers and trading technology that help APG and its clients to achieve its goals. It is part of our desk’s DNA to continually search for better ways to trade our clients order flow. For a trader, that is what best execution is about. With MiFID II coming into effect the focus on transparency around best execution has increased and many of the activities that we do in measuring, evaluating and changing our trading strategies become more important in 2018. Some of the activities will get an official status and we need to explain to our clients where and how we trade. For FX forwards, swaps and NDF this has become mandatory and although FX spot doesn’t fall under MiFID II regulation, we apply the same standards on our own FX spot trading.
Were APG quite early adopters of FX algorithmic trading and how important is making use of the latest platforms and technology for your trading and dealing activities?
At APG we have been using algorithms when executing cash equity and listed derivatives for over 15 years. We are very hands on when executing our order flow. We want to have control of the order and execute it ourselves electronically over our trading platforms. Here algorithms are an important part in the toolkit of our traders. APG was a relatively early adopter of FX algos. In 2013 we did a research study into the use of FX algorithms and after a few sample trades we decided to enable them for FX trading. We access the algos through our multi bank trading platform and trade-fills flow real time into our trading-management systems. While we continuously work on improving our technology and platforms, for APG it is important to use a matured platform and stable technology, not necessarily the latest.
APG has a global multi-asset trading desk
What types of FX algorithms are your trading team currently employing and what factors generally influence this?
We distinguish 3 types of algorithms with different urgency levels. Aggressive (high urgency), neutral and passive (low urgency) algorithms. Which type of algo you use depends on the trade-off you make between impact and opportunity cost. Most of the time, when we execute trades at APG, we aim for the lowest impact possible. During a trade we will manage our opportunity cost in a controlled way. This decision is taken because of our trading size which is generally larger than average. Our experience is that on large trades, if you trade over a longer period and trade smaller clips, on average the total trading cost will be lower. Therefore at APG we often use passive and neutral algorithms. It is the skill of the trader to analyse the markets, price action or other factors and to decide when to deviate from this default execution strategy. When there is a real high urgency of a trade we have the option to use more aggressive algorithms or trade on our multi bank trading platforms/ECNs.
Many large asset managers choose to utilise bank-generated algos as they believe it gives them the best access to the different trading venues that exist. What about APG. How do you source yours?
At APG we have strong relationships with many top tier banks that give us access to their liquidity and balance sheets. For FX trades we access this liquidity through two multi-bank dealing platforms, where we can do RFQ, RFS and use algorithms. Next to this we connect ourselves to two ECNs directly and trade via a prime broker. We are not planning to expand to more ECN’s. This can be done better, more cost effectively and with lower latencies by technology providers and our banks. We utilise bank-generated algos mainly for the technology they offer for structuring order flow and for connectivity to ECNs with smart order routing capabilities. We see the correct connectivity, quality of the smart order routing and policing of the liquidity providers as the main differentiators in choosing one algo over another. Next to this access the ability from a provider to tap into internal liquidity can be a differentiator. In that sense, we investigate how a bank’s desk E-desk is set up, technically and legally, what controls there are over order flow, how the Chinese walls work. This due diligence is important for our control over the order execution.
“We bucket comparable algorithmic executions across different variables and check if an algo is doing what it is supposed to be doing, compare the algos versus each other and versus different trading methods.”
How involved do you get in the testing and fine-tuning of the FX algos you are using?
APG likes to be involved early when new algorithms are being developed and we also continuously give our feedback towards the algo providers on what we find important in algorithms. With the initial testing we do not want to be involved as this is more for the developers, however when an algorithm is ready for live usage we do not hesitate to use new algos. Initial feedback is given directly to the provider and longer term TCA results are shared upon request if the sample size is adequate. In the end we would like to work together with our algo providers to develop algos most suited for our FX flow.
Are you using FX algos for every trade or only for specific situations, for example executing a larger block trade or dealing with more illiquid currencies?
APG mainly uses algorithmic execution when executing large orders. For smaller orders the added value is small and its often quicker and/or cheaper to use a multi-bank trading tool. For large orders we find it important to use passive algorithms that give us access to liquidity in the market. Our main goal is to minimize the impact of our trading activity. Algorithmic trading fulfils a role in achieving this.
“We see the correct connectivity, quality of the smart order routing and policing of the liquidity providers as the main differentiators in choosing one algo over another.”
What are the key benefits that you getting from algorithmic trading?
Important benefits we get from algorithmic trading are getting access to a large diversity of ECN’s and liquidity providers, accessing the internal liquidity of our brokers and their franchises and using the smart order routing and placement technologies that they develop. Next to this algorithmic trading also allows a trader to focus on more trades at the same time while still being able to slice large trades into smaller executions with smaller bid/offers and lower overall impact. This is important to APG as we have a global multi-asset trading desk where all traders trade multiple asset classes, often at the same time. In order to do this in an effective and cost efficient way, trading technology is vital.
Rigorous TCA is important to ensure that algorithmic execution strategies genuinely achieve a better result than more traditional trading methods. How do you tackle this and what sort of benchmarks are useful?
At APG we believe doing TCA is very important to analyse the different ways of trading and accessing market liquidity. Furthermore we think that systematic and independent TCA is needed for it to be useful. We have been using an external independent TCA provider since 2014. Our TCA partner provides the market data and together with APG develops benchmarks that we find useful in analysing our trades. For evaluating algorithmic trading we focus mainly on spread capture and information shortfall. We bucket comparable algorithmic executions across different variables and check if an algo is doing what it is supposed to be doing, compare the algos versus each other and versus different trading methods. This exercise if very difficult for many reasons, for example market data is still not freely available, volume data is not available and for TCA to be useful a large sample size is needed. After doing TCA on our trade activity for the past 4 years we are at the point where we can genuinely use TCA to make decisions in the trade process.
Post-trade TCA has great value of course but how important is the ability for you to see how an algorithm is performing in real-time and to be able to intervene if the deviation to a benchmark in terms of slippage has crossed a threshold that is too great?
For a trader it is very important to see how his or her trades are performing in real time. The degree of real time analytics and performance indicators that are available to FX in an integrated and centralized way is very limited. This is mainly due to the fact that the use of execution management systems (EMS) is not that common in FX. At APG we do not have an EMS for FX yet, unlike our equity trading where the use of a EMS has been integrated in the trading process already for a decade. We envisage that in the future we will integrate an EMS in the centre of our FX trading process and from this central point we will access the whole FX market, be it through ECNs, crossing networks, RFQ, RFS or through broker algorithms. Some algo providers do offer real time analytics for their algos. Although we find this a nice initiative, the added value for us is limited as we need it centralized over all our trading activity.
APG works for approximately 40,000 employers
There seem to be two schools of thought as to whether to let algos do their job or to go for a more hybrid, human-algo approach. What’s your opinion on that?
In the beginning we had a very trader dependant micro-management approach when using FX algorithms. Very tight limits were used and settings were changed frequently during the execution time of an algo. The hybrid human approach best characterized our usage of FX algos at first. Advantages were that trader skills were used and we could micro and risk manage the algorithms. Disadvantages are that you overrule the algorithms frequently and therefore influence the measurability and performance of the algorithms. In our case it made some of the TCA results on algorithmic trading less useful, while TCA is very important to show that algos add value and to decide which algos have the best performance and are suited with your trading goals. Therefore in 2017 we made some changes in the way we use algos, mainly that we use fewer and less tight limits and we always let an algo finish. This had a positive effect on the quality and fairness of our TCA results.
The correlation between the cost of one particular algo and the direct benefits it delivers is not always consistent. How big a challenge is that in determining which ones to use?
It’s a challenge also because of the fee structure. Until 2017 we paid the same fee for all our algos. In 2018 we are moving to a two tier fee model on FX algorithms were we pay more for the sophisticated ones that add the most value and we pay less for the more standard algorithms. Based on the workings of the algorithm, the depth and performance of the smart order routing and our own TCA results, we decide in which tier every algorithm falls. With this new model we want to reward banks that have employed algos that add the most value for us and encourage more development in algorithms that suit us the best.
As algorithms continued to be developed and deployed in the FX space, they are starting to appear very similar, almost like a commodity. How do you tease out and select the ones that are most suitable for your own requirements?
Some algorithms indeed appear very similar to the extent that we cannot tell the difference. One of our technology providers has recently invited us to participate in early testing of an FX algo wheel. This is functionality where a program will select from a list of similar algos with similar parameters a random algo that will trade the current ticket. Through the use of this wheel you will build a good database of comparable trades that in time will assist in choosing the best algo(s). This initiative originates from the equity world and we are enthusiastic to participate in this project.
What steps do you think banks and algo providers can take to increase the appeal of algorithmic execution and remove some of the mystery associated with it?
The biggest step for us would be that providers enhance venue analysis and show it to clients. Algo providers can give better insight in the venues and liquidity providers that they route to and show us performance data on these venues. Currently not all algo providers are passing on this level of detailed data. Also the workings of the order placement logic and smart order routing rules can be explained better. We find the policing and selection of the venues very important, but we don’t see how this exactly works and have little influence in this process. We would like to be more involved in choosing the right venues that suit our orderflow. Our equity desk has performed a due diligence qualitative study on all its algo providers. Through 20+ questions they gained insight in the inner workings of the algos per provider ranging from smart order routing, machine learning to how the algo accesses the systemic internaliser. They use this qualitative study next to their TCA to choose which algos to use. For FX we want to roll this out in 2018.
Do you see any risks in using algos or conflicts of interest that might arise between provider and user?
We see one main risk that exists in the use of FX algorithms between the provider and the user, being the potential misuse of information that the user gives through using the algo. FX algorithms are sold as an agency solution in a market that has been characterized until recently by a principal trading model. Providers offer algorithmic trading technology with connectivity and smart order routing for a fee. When the providers do only this, the agency offering works and the FX market can use this model next to principal trading. The providers being mainly banks should take all efforts to prevent any misuse of information of flow that is being handled by the agency algo desks, with segregation of duties and Chinese walls. Providers also offer this agency solution with access to the banks principal liquidity. Potentially this can be a conflict of interest if managed the wrong way.
Are you likely to expand your use of algos?
APG currently uses FX algorithms for 40-60% of its FX flow and we have no plans to increase this in the short term. The development and performance of our current and potential new algorithms will determine if we will use it more or less. If we compare the use of FX algorithms versus other assets classes we trade like equity and futures markets, there is definitely room for more algorithmic trading in FX. That said, in markets like fixed income and interest rate derivatives, there is the most scope for increased electronic and algorithmic trading.
