Measuring cross-asset transaction cost analysis (TCA) and trade performance still faces major constraints, says Baillie Gifford’s Petros Kyliakoudis, with unique nuances and complexities across varying asset types.
Speaking to Trader TV, the head of trading research and analytics discusses how his buy-side team approaches these challenges and evaluates cost and quality of execution for equities, FX, and fixed income. Kyliakoudis also shares how Baillie Gifford has adapted methods for feeding information between portfolio managers and traders, and unpacks the major lessons he has learned from his firm’s TCA and algo usage this year.
Interview
Josephine Gallagher – Measuring trade execution costs and performance can be challenging in volatile environments, and we’ve seen plenty of that this year. Characterize some of the limitations to measuring TCA (Transaction Cost Analysis) and quality execution across assets?
Petros Kyliakoudis – Well, first of all, I would focus on equities, which are the more established ones and easier to discuss. We actually see day-to-day execution, use of algos, crossing networks, way more advanced analytics about multi-day orders, or even investment analytics with investment behavior clustering and profiling. On the other side, you see execution analytics in the millisecond precision, about like spread cost and peak sizes. One of the challenges for equities is that we cannot really compare day-to-day execution with multi-day orders. For example, if you trade for 10 days in a row, you cannot really measure it from the market arrival, because five basis points from arrival might mean nothing in a 10-day order. So there are some limitations around that part, and this feeds through the models. The models are usually trained for day-to-day execution, not for the multi-day orders. This leads to discrepancies, so we need to understand the limitations of our metrics and the analytics we have available and use them appropriately. Now, if we go to the other asset classes, FX as an example, we use algos, we use RFS, differentiation types. There, the analytics are way more advanced compared to fixed income. As an example, in fixed income, you don’t have a mid-price. What you have is an estimate of a mid price, and what that means is your results will be good on average, based on the law of large numbers; you are assuming that the results you have will be correct in the long term. So these are the limitations, I would say, that are across asset classes.
Josephine Gallagher – Now every firm does it differently, but how does your team evaluate TCA and performance?
Petros Kyliakoudis – I think we need to understand the different perspectives across asset classes and across different levels of analysis. Let’s start with equities. Equities, you have the day-to-day execution, and you have the multi-day orders about the fund manager’s perspective. So for the day-to-day execution, you can easily look now the quality of execution through a new tag we have. The tag five to seven, which allows us to understand which algos are used for every execution in fill by fill basis. And we have the millisecond precision timestamps from our brokers, so we can actually now drill down in each algo, isolate it, measure the performance, and understand the impact they have. In some scenarios, we see four or five algos for one trade, which means that if you don’t slice it down and understand the impact of every algo in specific (detail), you are over or underestimating the execution quality. So we actually now have a proper scorecard using a multi-benchmark approach, where our traders can access and decide what is the best strategy based on their parameters. It is very different when we discuss the fund manager’s point of view, where they will only look for multi-day results, and the algos are always under the hood for them. However, using the best algos can minimize the days to completion and minimize the impact they have, even in the multi-day case. So in that sense, they are interlinked, but we need to remember, actually, to use different lenses and different perspectives in the analysis. The trading cost is one component of this analysis. We also look for investment timing. We are looking for different levels of AUM to understand what will be the market footprint of the trading. We are looking at metrics like success rate and hit ratios or payoffs to understand what we could do differently or what we’re actually doing well. By doing that and putting this in dollar terms, we are making it more personal for our fund managers, which in turn makes it actionable. So that is for equities. On FX, we are using a lot of RFS for our spot rate. However, recently we see the increase in use of algos. But it is counterintuitive from the equity background, because here you’re using the algos in very big order sizes, as opposed to the smaller sizes on equities, because on the FX you want to capture the unique liquidity the banks can provide in their books. In fixed income, definitely, we see the platforms increasing their functionality over time. However, they are fragmented so we don’t have correct pricing over time. We know from the market discussions that a lot of the liquidity depends on the counterparties, so if you have a good coverage, a good dealer on the other side, you take the liquidity. In fixed income, liquidity is the key, let’s put it like this,
Josephine Gallagher –How has your firm adapted that feedback loop between the pm and the trader on the
Petros Kyliakoudis – As I mentioned, we have a separate analysis for our investment teams. One part is to understand what is the impact of changing the order sizes and that is related to the maximum level, or the minimum level of AUM that can achieve without being detrimental to the performance. We are listening to the IRC, which is the Investment Risk Committee, which oversees these analytics. We have a separate team called Investment Analytics, which looks at the data for the investment teams from the traditional risk perspective or the back office. We now have the trading analytics team providing feedback about how trading is impacting the performance, and then we have six-monthly meetings with our investment teams where we actually discuss the results, we are showing them outliers and trying to find better ways to communicate, to also improve the timing, to improve certain behaviors, things like biases. People used to sell the best-performing stock, to buy the new ideas. Is this the good thing to do? If you don’t have the data, you will never be able to say to your fund manager that, while with our analytics, we can actually do that and give them very concrete feedback, which makes it personal.
Josephine Gallagher – Finally, what are some of the lessons you’ve learned from studying TCA or your algo usage this year?
Petros Kyliakoudis – At the beginning of Q2, you had “Liberation” Day. So we saw the markets becoming very volatile, especially in the US. Everyone was expecting trading costs to increase through the roof, which happened for a couple of days, and then everything returned to normal by May. One lesson from this is that markets are more resilient than we make them to be, and you need to remember that volatility is not always bad. Volatility can create opportunities if your investment goal is to capitalize on that. And second, when you decide your algo, you should let it run its course. If you’re changing the benchmark throughout the execution, you are usually damaging performance. So let the algos do their job. They are trained, and we should trust the analytics and we should trust the process which, at the very end, will actually improve the efficiency and the quality of the execution.
Catch Trader TV shows each week:
To watch Trader TV This Week shows go to – Video
Or make sure to follow us on LinkedIn – Trader TV’s LinkedIn
Available as a podcast – Link
