Andy, please tell us a little about your job and responsibilities within Invesco.
“We have often found given the right conditions that you can see vastly improved results versus risk transfer when using algos in markets which traditionally have low primary market volumes or are deemed to have poor liquidity”
Based in London I am a member of the Alternatives trading desk, part of Invesco’s Global trading team. I am responsible for executing and managing trades on behalf of our fund management groups, across a range of fund types, strategies and locations.
The Alternatives desk facilitates multi asset trade execution, with a heavy focus towards listed and OTC derivatives across Global Equity, Fixed Income and Commodities and my main centre of attention Foreign Exchange to include cash and options. Outside of trading we spend a lot of time looking at the market, interacting with our key stakeholders and counterparts and analysing our capabilities & performance. In terms of trading we use a combination of traditional voice trading, e trading via RFQ and algos and manage the workflows via OMS and EMS
Broadly speaking what are the day to day objectives of your dealing team?
Simply put, the aim is to enable consistent high-quality execution outcomes for our fund management groups and their clients. Invesco views the dealing team as part of the investment process that acts as a conduit between the firm’s portfolio management teams and the market so the dealing team must have a clear understanding of the investment objectives of the investment teams it supports. Clearly no two markets or products have the same profile so having a strong understanding of market structure and liquidity dynamics allows the team to employ effective implementation strategies.
How hands-on do you like to be when executing your order flows and how much importance does the firm place on continuously working to improve its technology stack and trading platforms?
Certainly, given the nature and increasing volume by number of our flows and the challenging liquidity profiles of some of the markets we can be engaged in, it can never be a set and forget mentality across any of our trades. For FX specifically, this includes adopting an electronic RFQ wheel, which use live market data and historical execution data to systematically execute trade orders. This allows traders to spend more time on large and complex trade orders using pre-trade TCA tools to evaluate market liquidity. We are also looking to build monitors using live and historical market data to provide dealers more information, so they can make more informed decisions. Invesco Global Trading have committed a lot of time and effort in the technology and research space, and we are lucky to have a dedicated in house resource for trading research that sits within the group. As we continue to gain access to cleaner market data on our execution outcomes and performance as a result of the evolution of our internal TCA, we can be better informed and increasingly discover opportunities to improve our trading styles and our platform setups which in turn allows us to be more efficient and consistent with our output. All of this forms part of a team effort to build a world class trading platform that can adapt to these fast-changing times
In what ways have new regulations and the arrival of best practice guidelines like the FX Global Code influenced how you go about measuring and evaluating your own trading strategies and need to explore new execution alternatives and ways to undertake them?
Best practice guidelines such as the Global code are additive to institutions like Invesco who are in many terms a liquidity provider to the market yet still a price taker. The global code allows larger buyside firms to become more comfortable playing that role of market maker via FX algos and in many cases now gives us the opportunity to curate our own liquidity pools either pre-determined or bespoke. This certainly gives us the freedom to test and explore with less fear of encountering adverse behaviours.
When did Invesco begin to explore the use of FX algorithms and how much reliance do you now place on them?
In Europe as bandwidth and understanding has increased across the desk it has been a steady and important growth area for the desk over the past 3 to 4 years from a relatively low base. Algos are another important tool in the toolbox for traders especially as FX trading has become more electronic over the last decade and, as a result, the market has become more fragmented. For example, this year due to COVID-19 the market experienced periods of very challenging liquidity. As a result, the use of Algos became very important for trade executions of large trade orders and orders in illiquid currency pairs.
Invesco’s London office
What are you particularly looking for in terms of the functionality that FX algos can offer?
When executing any transaction there are a number of things I would consider – urgency and nature of the order, size of the order, time of day, current general market dynamic, the expected bid-offer and expected liquidity profile of the currency pair. If I have taken the decision to trade via an algo then ideally, I want to be able to add some of this thought process into my order. So key features would be the ability to dial up or down speed/aggressiveness which goes in part hand in hand with the optionality to choose the liquidity profile. Dependant on the provider and chosen algo you may be doing the latter as a result, for example internal strategies with differing degrees of aggression. A major component of any algo for us also has to be the ability to interact, pause or adjust in flight, set limits and increase/decrease participation at certain levels – the ‘I would’ scenario for example. It can also be nice to be able to switch strategies in flight and have a good read of what the market looks like for your trade or remaining balance in terms of spread and the expected cost of execution. These are things which we already see from some our providers.
What sort of FX algorithms are your trading team currently employing and what factors generally influence this?
The three main styles we concentrate our efforts on are a Passive, Dynamic or a TWAP plus type algo. As I mentioned, with any execution there are the general considerations around current market conditions which influence the decision – market impact, liquidity, expected costs from spreads, perceived market risk, time sensitivity, time of day or any pre-determined instructions from the fund manager or client. Where possible, given time and when operating in a liquid market with good volumes, I would always look to try and capture spread and act broadly passively. However, there is a clear trade off between taking too long to execute and assuming too much market risk so we try to use pre trade tools, expected spreads and market monitors to aid that decision. As we have clearly seen this year the condition of the market is volatile and so we have to remain ready to adapt our approach almost continuously, often in flight, and ensure we are always trying to learn from each outcome.
To what extent does an understanding of market dynamics and liquidity inform the decisions you take on how to deploy an FX algorithm and the execution style that may be most appropriate?
This is an interesting one as the answer isn’t always as you would expect – high turnover/low spreads equals good for algo, but undoubtedly something that we consider across the life of a trade and any individual algo execution along the way. For example, we have often found that given the right conditions, you can see vastly improved results versus risk transfer when using algos in markets which traditionally have low primary market volumes or are deemed to have poor liquidity. Again of course we have to be careful to caveat that there is always the market risk trade off and it can really depend on the profile, size of your order and the time of day.
Earlier this year as spreads widened, due to poor liquidity in March around the COVID led market disruption and subsequent equity sell off, the market actually saw a heavy increase in FX algo usage as participants took preference to algos over offloading their risk via more traditional voice and e routes. This would certainly be in contrast to previous risk off scenarios, where we have seen heightened volumes coupled with an increase in volatility of pricing, spreads and depth of order books leading to buyside participants increasing voice risk transfer execution, versus working orders with trading desk or using RFQ/algos. To me this demonstrates that the market has become more comfortable using FX algos which fits the trend in YOY volume increase, although it could also have been impacted by the move to work from home setups. It should be noted, however, that looking at performance it may not necessarily have been the right avenue of execution during the initial period, as generally participants underperformed risk transfer despite wider market spreads.
So to chose the correct algo and execution style it’s critical that you also have an understanding of how and where your chosen algo is working the order – Lit venues, mid books or internal market and are you potentially showing a skew to the banks franchise?
Market dynamics are also key. All pairs have differing liquidity profiles over the course of the day so you should be asking: how your order compares to average daily volumes at that time, how much trades on the primary market on a given day, how much flow is deemed to be internalised and who are the key market participants. Pre trade TCA can be important as is having a good view on expectation versus your chosen benchmark. All that should feed into your decision-making process and strategy choice.
How has the way you use FX algos evolved and how prepared are you to let them do their job and avoid micro-managing various parts of the execution process?
I think when FX algos were first introduced the idea and excitement was around being able to access the lit markets and to act as the old spot trader of a bank would have done when passing a voice risk order or working an order for you, but whilst maintaining your own level of control and having more direct ownership of your expected outcomes. So the focus initially and availability for me was to be using more dynamic algo’s, working to capture spread but taking liquidity when itwas there and aiming to limit time and market risk to complete orders. As usage and offerings have evolved the FX market has seen a large increase in internalisation and as a result, lit venues have seen market share drop significantly. This leads to an increase in banks offering internal style algo’s which work against bank franchise flow or access market mid books, rather than posting interest on the lit venues. This has certainly been the recent shift over the last two years for us but dynamic algos also remain relevant. Although the number of algo styles we use has not dramatically increased, the number of parameters within each we can interact with has, and therefor requires more consideration.
It remains important to constantly monitor an algo order given that we know markets can change in a very short space of time. The long-term mindset would be to be in a position to have set defined parameters, consistent across providers and strategies, so we can further enhance our performance analytics and have a clearer picture of which algos are performing for us and where traders in their decision making or banks individual strategies are adding alpha.
What are some of the key things you are trying to achieve when using algorithmic FX trading techniques?
For us reducing market impact is an extremely important consideration, for example we can often be working trades which are relative value and not necessarily in the same asset class that can take place over a number of days or even weeks. Algos can give us the ability to work within a set time frame or work quietly and hopefully capture spreads when trading correlated assets. At the same time, over the course of our execution year, we have an opportunity to compress our trading costs, improve entry or exit levels for our clients and behave in a consistent manner. It is never our aim to be an extreme percentage of market flows on a given day and using algos can also allow us to ensure that we are not artificially pushing the market or holding it up. In turn we can capture that data and feed that into future development and practices.
Best practice guidelines such as the Global code are additive to institutions like Invesco
who are in many terms a liquidity provider to the market yet still a price taker
How do you source the FX algorithms you use and what sort of information about their capabilities and performance attributes are you seeking from providers?
At present we are using algos which are provided by our major trading partners and are available within the FX platforms. These are integrated with our order management systems. That gives us a good balance of maintaining our focus towards STP and using technology to give comfortability around control, access and safeguarding and does not impact our ability to access the main suites of algos the banks provide. Having access to all of the leading bank provided algos allows us to see what works and fits our trading style. We look for information around how the orders are worked, for example, if we are working passively – who is seeing that, and if it’s an internal style algo – what does that really mean to that provider and how does it interact with the E Book, spot or options desks. Generally, we want a high percentage of internalisation of flow but not at the expense of either, adding risk to the banks e-franchise to a point where there can be a spill over into the primary market or where we are showing skews to predatory participants, which can cause spreads to move against us without seeing market flows.
What do you see as the main benefits your dealing desk is getting from using these toolsets?
Using FX algos, pre trade tools and diving into the TCA performance after the event has really helped us gain a broader understanding of the environment in which we are operating. The benefits from having that information at hand should enable us to provide more useful insight to our stakeholders and ultimately their clients. To be able to give realistic time frames around order flow, expected costs and the ability to add or reduce risk. It’s key to understand that if we need to get things done quickly we can, but conversely if there is another opportunity to improve our execution if we have the right conditions. Ultimately this should help us to fulfil our goal of achieving consistently high quality outcomes for our group.
Why is trying to compute and obtain accurate execution metrics in FX still such a challenge and do you expect this to become easier before long?
Given the fact that there are so many liquidity providers, venues and execution platforms available for the market, and banks differ in their approach to each one, coupled with the liquidity profile and dynamic that is not consistent, this presents the challenge to establish firm tradeable liquidity and an absolute picture of the market. In many cases banks will stream tighter pricing than primary and secondary markets and due to increased internalisation this can be a trip hazard when trying to compare volumes and spreads in those venues, versus algo or RFQ execution. Seeing this clear picture of the truly available market remains at best estimated and while access to this information has improved, real time market spreads monitoring, volatility of spreads and tick volatility is generally backward looking. I would expect this improves as we see more engagement in micro structure reporting across FX and if our counterparts are looking to reduce the number of venues and liquidity providers they are plugged in to. My hope would be that this leads to top of book spreads being more consistent for our client type, with firmer robust liquidity which we can analyse effectively and be confident of having consistent results against.
What sort of data are you looking for from the TCA and other analytical techniques that you use to help evaluate the effectiveness of FX algos?
It’s always nice to know if you performed well versus the benchmarks, so visually I like to see performance in basis points or in terms of lead currency P+L as a quick guide of how well things have gone. I like to see Participation rates versus primary market or the estimated market volumes during the period of your order, and how that compares to historic. The other key metrics I would look at are internalisation percentage and passive/mid/aggressive fill percentages. All of those should give you a good picture of your order. You can also choose to look at spread savings or spread earned, but I find that somewhat counter intuitive if you have performed badly versus the risk transfer.
What’s your opinion about the relative merits of using independent third-party TCA solutions as opposed to analytical services offered by algo providers?
I think it’s certainly healthy for the market to have third party providers providing TCA because it allows you to drill down on the algo performance, not just the performance of the algo versus the market and separate out the two. You may well have beaten your internal and the banks relevant market benchmarks but did the algo perform as well as it should have in that period. Long term that keeps everyone motivated to improve and that can only be a good thing.
We have mentioned before that the correlation between the cost of one particular algo and the direct benefits it delivers is not always consistent. What are your own thoughts about that?
If I understand correctly, I would say if it is clear that the cost of a particular algo outweighs the derived benefits or if it is a scenario where its cheap and performance is negligible, then if you are happy with performance of your other available algos and the relevant associated costs, you should not be using them or using them as a negotiating tool to drive down costs for your other algos.
What would you like to see developers of FX algos and analytical toolsets focus on which would improve their functionality and strengthen their value proposition for you even further?
This comes back to the question around understanding of market dynamics and liquidity. For me we need to focus on transparency. Around how our orders are performing in flight, recognising flows which have gone through the market during our order and allow us to target participation rates more accurately or see where we are potentially not getting fills. I would also like to see that if our trading partners are able to consistently outpace primary markets, by recycling their risk via franchise flow and achieving very high levels of internalisation then, as key participants and trading partners, it would be good to also have that transparency and opportunity when working alongside them.
Do you expect to see Invesco’s use of FX algorithms increase in the future and what factors might help them to gain even more traction amongst firms like yours?
Of course, dependant on market conditions and on algo performance moving forward, I would be confident that our usage of FX algorithms will continue to increase and I would expect to see automation for our smaller orders become a priority as a higher percentage of our flows fit into the algo route bracket. Less traditional pairs such as NDFs should see increased volumes, active venues and liquidity providers available which could also be a potential area of for us and the market if fragmentation concerns abate.