How long have you been using DMA and algorithms for trade execution?
We started about two years ago. The main incentive was that it gives us better control and visibility of our trading activity, which allows us to react very quickly to market conditions. That is a notable improvement on sending trades out to brokers for execution and having no real transparency on progress. Particularly when a broker might be running multiple programs across numerous names, this wasn't a very efficient process. The switch to using DMA and algorithms in-house has definitely had tangible benefits in terms of performance.
How does the trading function engage with ABP's fund managers?
All the fund's front office operations are located in an office at Amsterdam's Schiphol airport; in fact the fund managers are on the same floor as the trading desk. Therefore when a manager has a new strategy and trade list to discuss with us, it is a very straightforward matter. It is a consultative process, in that we are able to give managers a clear idea of the likely execution profile of particular names in their list, whether executed directly by ourselves or by using algorithms.
Are you using broker algorithms or ones developed in house?
We are using broker algorithms, but many of them have been customised to our requirements. In view of the data management and maintenance workload that would arise if we wanted to develop and maintain our own in-house algorithms this simply wasn't practical. Therefore the quickest way for us to go live in the market with algorithms was to use those provided by our brokers. As regards the actual front ends, we use a couple of broker systems and also some multibroker interfaces, which allow us to disguise our flow.
Is there any front end that you use in particular?
One of the main systems that we use is Goldman's REDIPlus. It had
been available in the US for a while when we first started using
it, but I think we were one of the first European based trading
operations to adopt it. As a result, we were able to have extra
functionality added in terms of displays etc, which was valuable.
Goldman has also done a significant amount of bespoke work on the
algorithms within REDIPlus for us. This has continued to evolve,
so additional refinements we request for both algorithms and the
interface are still being made.
ABP Headquarters in Heerlen
Who provides your pre and post-trade analytics?
We have developed all our own pre-and post-trade analytics in-house and have been using them for the past three years or so. We built the pre-trade analytics ourselves on the trading desk. We feel it is the most even handed method, otherwise you end up using one broker's data to predict what another broker's algorithm might do, which is unlikely to yield comparable results. It also gives us far more flexibility.
As regards the post-trade analytics, we have found the big advantage is that we can do the analysis straight away, rather than having to wait for data that might also be incomplete. It also allows us to analyse by broker, market, and individual stock across any timeframe, so we can quickly detect where any of these elements are costing us money.
How do you apply the pre-trade analytics to your trade lists?
We will load the list, which may contain a thousand names or more, into the pre-trade analytics which split the stocks into three main groups:
- Those that represent a large weight in the list
- Those that are likely to prove problematic
- "Vanilla" stocks that because of their trading characteristics and/or size are relatively easy to execute
The first two categories will require individual attention, while the last group will typically be program traded using algorithms. Generally we find that the pre-trade analytics do a good job of segmenting the various stocks in terms of execution difficulty. However, we will still review the categories by eye for any potential problem stocks, checking things such as spread, volatility, sector, weight in basket, country etc.
Do you use your pre-trade analytics only for stock screening, or do you use them to select the best broker algorithms as well?
In practice we find that for many categories of algorithm, such as VWAP, there isn't a great deal to choose among them all. Yes, some are smarter than others, while others are market or stock specific, but over the long term the average performance difference is fairly small.
It is much more important for us to understand the algorithms well enough so we know what constraints to set around them and when to turn them on or off. That (and exception monitoring) has a far greater influence on execution performance than the choice of algorithm.
Do certain members of the trading desk tend to specialise in certain types of trading?
Yes. In general, I deal with most of the programs, which represent 80-90% of our total flow, which involves tracking the "vanilla" list and any exceptions that may arise within that. At the same time, some of the other traders will focus on the individual stocks, possibly even following just a couple of stocks for the entire day. Somebody else may just focus on hunting for liquidity, trying to find indications of interest (IOI) etc.
That specialisation applies to the technology as well - each of us will tend to focus on perhaps one or two of the systems that are best suited to the type of trade we usually execute. Nevertheless, we all have sufficient breadth so that we can cover for each other when required.
What geographical range of markets do you trade and what differences do you notice?
We are global, but the US and Europe are by weight the largest areas of exposure. They are also the regions the trading desk tends to focus on, while trades for Asia are usually handed to brokers for execution.
In terms of differences, the US market is obviously heavily geared towards the use of algorithms - you only have to look at the bids and offers with no more than 100 shares sitting on each to appreciate that. Under those conditions you are effectively obliged to use algorithms just to feed in multiple 100 share slices to get the trades done.
By contrast, in Europe you can still find the larger blocks of stock that are more elusive in the US, but on the other hand the spreads are wider. While we use algorithms both for US and European stocks, I would say you need to exercise more care on the European stocks because of the wider spreads. If you just turn on an algorithm that feeds in a slice every x minutes and pays a 30-40bp spread every time, then you can end up paying away a lot of money. Particularly where you have a stock that may trade in a small range with only four or five prices printed on the day, an algorithm is simply not as smart as a human trader who can sit and wait on the bid and/or go looking for IOI or a block.
There's been quite a lot of talk lately about incorporating newsfeeds directly into algorithms' execution logic. Do you have a view on this?
I can see where this could possibly work as regards aggressive algorithms, where you might have a major news item that would suspend the operation of that algorithm automatically, but that is something you would be doing yourself anyway, with popup notifications etc. Beyond that, I can't really see a huge amount of practical use for this, and we haven't heard much about it from the brokers as yet.
However, a related area where we have seen some activity is in incorporating major levels in stock indices into algorithms. For example, if you were using an algorithm to trade a FTSE stock and the FTSE index broke through 6000, the algorithm would take this into account when executing trades.
One of the ABP trading desk's main tools: Goldman Sachs REDIPlus
How do you feel buyside DMA and algorithmic technology compares with that used on the sellside?
As I used to be a sellside trader until about three years ago, this is of particular interest to me. On a recent trip to London, I visited several of our brokers to see how their technology had moved on. From what I could see, there was very little difference between what we have on our desk and typical sellside systems. I think this is interesting in that it shows how quickly the latest technology is becoming accessible to all categories of market participants and not just those with the largest technology budgets.
However, as I mentioned earlier, we have come to the conclusion that it is not the algorithms and tools you have but how you use them that counts. While you hear plenty of claims by various brokers that their algorithms are better than somebody else's, I don't think there is actually an enormous difference among them.
While it usually receives far less attention, it is ultimately the algorithm selection and monitoring that makes the real performance difference.
Do you see the role of DMA and algorithms playing a wider role in your trading?
Our treasury department have already extended their use of these to include futures for the markets they are responsible for, and we will be following suit. We already have the technology pretty much ready to go and so we will be implementing DMA and algorithms for stock index futures soon.
What are the main advances in advanced order execution you anticipate in the coming year?
I don't see that there will be huge changes; I think the marketplace has reached the point where everybody is talking the same sort of language. Therefore, I don't feel we will see quite such a fast pace of innovation in the near future as we have in last year or two - it will be more a case of people catching up.
Having said that, I think there is definitely scope for enhancements in areas such as sourcing liquidity. Tools that would facilitate the flagging of IOI and the finding of blocks (especially in the US market) would be most welcome.