By combining historical and current data and PM trading style to optimise executions, Mark Kuzminskas, Director of Equity Trading at Robeco Investment Management has been able to make significant savings.
I wanted to devise an optimisation platform whereby we would be able to understand how algorithms performed under various circumstances. I was attempting to build a grid based on the performance of these broker algorithms so that when an order came in, we could look at the grid and determine which algorithm would be best suited for that particular order.
Moving on a couple of years, I was introduced to our new transaction cost analysis (TCA) vendor, providing daily, quarterly and annual review information, and during discussions with them I found out that they had actually already devised an optimisation tool.
When I started out in this industry in 1990, I had to learn the language of the business; we were trading in a fractional environment, but liquidity requires a lot of work. I wouldn’t simply put my order into one algorithm and leave it at that; I would use different algorithms at different times and that was the genesis of the optimisation platform.
The optimisation platform we developed and deployed routes out in a directed DMA fashion to the designated liquidity centre. It makes a determination as to whether the order that it sliced off at a given point in time from the parent order is going to go into a lit or a dark market, whether it is a limit or a market order, and how to react to the volatility within the bid-ask spread. This data is combined with the portfolio manager’s (PM’s) data — the stored trading investment profile with real time analytics. That is the key to this, i.e. marrying those two elements.
When I think about trading, the more that we cut out the middle, the intermediary and the process, the better the results. When you think about best execution as the best price given the liquidity characteristics of the stock, market conditions, relevant news in the marketplace and most importantly the PM’s objectives, the optimiser provides us with the PM’s profile and objectives, because each PM has an investment DNA. So we are coupling that with the market characteristics or conditions out there in the industry. Given the volatility in the marketplace these days, a human trader can only trade five to ten orders at a time effectively.
For example, it is 11:00am and I’ve been working an order for an hour and a half in the optimiser. The optimiser can see the volume in the stock over the last few days and can compare it to how it is today. It can look at the volatility of the bid-ask spread, the size of the bids and the offers, the overall performance of the stock relative to its peers, to the market at large and it can do this on a real time basis. A human trader can look at various factors to give an idea of the position, but cannot do it in the same way as a machine.
The optimiser marries the PM’s profile with the real time analytics in a way that feeds the order to the marketplace accordingly. We use broker DMA pipes to get to the liquidity centres to trade but the decision-making in terms of the routing is done by us using the optimiser. As you can see we have taken out many of the intermediaries throughout the process.
Over the years of using TCA I’ve found that the more we remove the intermediary layers, the better the performance. My advice to PMs is that they can have really great performance if they implement trades themselves, because for example, they really know how urgent an order is when they say it’s urgent. And when you consider volatility in the market on any given day with regards to executing a particular order, the PMs know how much sensitivity around price versus volume they truly have.
The advantage of the optimiser is that, although there is a fair amount of probability associated with superimposing their viewpoint into the real time analytics, it is a pretty accurate guess and is fully automated. By balancing out the positive versus the negative trading environments methodically, we have already improved our trading by 20 basis points. We have developed an execution blueprint for various types of order profiles and we have been able to engage with traders in such a way that they are brought into the process.
Typically between one and five orders account for the lion’s share of the costs. Instead of focusing on the tails that can move performance significantly in a positive or negative direction, what we have focused on is orders that are one to two standard deviations under the curve. It is by trading those orders in a much more methodical manner that we have been able to reduce implementation costs by up to 20 basis points.
What this does is free up the trader to focus on the trades that are in those tails. While it may be a handful of trades, they end up accounting for the largest costs as the tails are so big. The optimisation tool put us in the position of the PM as a trader so that the PM doesn’t actually have to do the trading. There is a fair amount of customisation associated with this. It’s not just an ‘off the shelf’ package. You have to develop the system because the optimiser learns more each day about the trades and continually improves the process. The more data you supply it with to develop a benchmark or baseline benchmark the better it allows you to know how to develop a trading plan where you’re having trouble. The quality of data is important particularly behind the scenes in terms of ensuring you have the proper data fields which come from the order management system into the main system as part of the overall baseline of analysis.
Data management issues
The processing is dealt with by S J Levinson, but we look at the real time tick data which has to be fed into the process. There are many hours of number crunching and processing each night in order to get to the starting point the next day with another day’s worth of implementation or execution knowledge that we can add on — so we are building layer by layer every day. Once we had developed a baseline we were then able to set goals for the various strategies, order types and put the plan into action.
There are two factors to consider. One is that from a practical perspective we don’t have to put an order flow into the optimiser on a given day so it’s at our discretion. That’s part of where a trader’s role is so important particularly on days like the Monday of the Boston bombings. But also as we have a baseline execution benchmark, we now can look at the job a trader does when he/she pulls an order out and trades it differently to our benchmark goals.
To illustrate, based on all the research available and the industry commentary, I didn’t put those orders in the optimiser, instead I traded them myself. Through our analysis it became apparent that the rate at which I was trading was not the rate at which the optimiser would have traded — so we can compare my trades relative to what the optimiser would have done.
It was then easy to determine whether I had added value or not. Those particular trades did add value, but there will be times that we can see when I might not add value. For example, say I have an order to buy in a down tape and I think this is the time to go all in, remove the trade from the optimiser and just go for it myself and it turns out that the market is in a downward spiral for the rest of the afternoon — it’s obvious that I executed too quickly. The optimiser would have flagged this up.
I know that there are desks that utilise TCA. It’s part of a compensation discussion for their traders. I’ve never been a believer in TCA as a means to benchmark traders on their compensation. But now that we have a plan based on the benchmarking of our historical performance and how we would expect the trade to develop, that when we go away from the optimiser — there are guidelines that help you analyse how a trader is performing.
As a relatively new platform for commercial use we are able to provide regular customisation around the optimisation that we are looking at, so we have been able to put things like kill switches into good effect.
If we see that certain things are going awry in the marketplace, we can hit one button and it shuts everything off at once. We have a warning system on the platform that alerts us to unusual activity. We also have a messaging system that is activated when something doesn’t appear to be executing according to plan; for example if the optimiser stops executing then we need to know about it. If it’s executing above plan we want to see why that might be the case. So there is some potential for development there when a product hasn’t been vetted by many clients to the fullest extent.
We are the risk control so we are building certain benchmark measures into the platform to alert us where necessary. There are some measures that are automatic in terms of notional limits — order size limits for example. Even when we put a market order into the optimisation platform, there is a calculated limit to provide an execution band so we don’t have a ‘runaway train’. There are risk control parameters embedded within the system but we have also added additional parameters so that the trader can take more responsibility and we are able to customise these parameters according to our risk tolerance measures.
Looking back, looking forward
I wouldn’t do anything differently I don’t think; I’m very comfortable with the solution that we have. Because we adopted the system early, we are very fortunate to be working with a provider who is enthusiastic about adding functionality that assists us and which would be beneficial for future users. So it’s a good working relationship from that perspective. In terms of further development, each portfolio manager (PM) has their own style. If the optimiser knows which PM it is dealing with and what their historic profile is, the optimiser will change the implementation profile for each PM.
So you could take the same stock for two different PMs with two different profiles. One is the kind of PM who is ‘legging into it’, who might recognise that its time horizon could be extended a bit further, whereas another might realise it has got to tighten up the horizon and really trade more quickly and look for liquidity in a slightly different way.
This is what makes our business unique. As traders we understand why we organise our desks to trade for specific PMs — so you get to know their tendencies — but it is actually difficult to adhere to that consistently because of the human element. We are emotional, we may lose sight of the PM’s DNA and we get caught up in the moment and the market dynamics at any given point. What we learn however is to let our winners run but to cut the losers more quickly.