ORDER PROFILING.
Learn from the past to improve your returns
Vlad Rashkovich, Global Product Manager for Trade Analytics at Bloomberg explains how buyside traders can rip sizeable benefits by using a top down approach in the analysis of their order flow.
You have your EMS connected to brokers, news and orders are flowing through. You monitor the slippage of each order against various benchmarks in real time. You might handicap your performance with a cost model. You’re paying for all that, of course. It’s the cost of doing business. Right?
Right. This morning you got a number of difficult orders. How does your setup help you to win them? You look at the market, at the news, you size your order versus the Average Daily Volume. A good pre-trade tool with a solid cost model and a volume estimate can definitely help. You couple it with an IOI flow and you’re up for a good start.
Wait a moment. What about all these orders you executed over the years? Don’t they have some information about portfolio managers you work with, perhaps their patterns and biases? What about brokers and algos you have used, and which worked better for which type of order? Shouldn’t this history tell you, at the very least, about your own trading style and potential blind spots?
Another aspect is venue analysis, which has been very hot recently. With the proliferation of execution outlets, high-frequency players and dark liquidity, it’s natural to be concerned and want to control where your orders are being sent. As a result, the buyside demands increasing transparency from brokers. An urge to make some changes in algo routing logic is understandable. It is also relatively easy to implement.
Portfolio Manager (PM) profiling
A trade usually starts with a portfolio manager, and the largest improvement could come from analysing PMs and trading strategy alignment to the short-term momentum implicitly built into the orders. If the trader is wrong on the expected short-term alpha, he can try to speed up and find the best algo and venues to help him, only to realise that the right way to handle this order was to slow down. Venue analysis, broker selection, dark pools, IOIs – nothing will help if the trader’s overall strategy is not aligned with the portfolio manager’s.
A case in point, you work with a growth portfolio manager. Her “buys” might be momentum driven and require aggressive executions. Her “sells”, however, might be driven by the dips, where the momentum seems to be vanishing. If over time these dips are statistically significant to point for reversion, then the sell orders might benefit from a slow execution.
Real life and behaviour is more nuanced of course, thus the same portfolio manager might respond differently to various market conditions across countries, sectors, volatility and many other factors. As a result, you’re facing a multi-dimensional analysis where advanced machine learning techniques could decipher the signal from the noise.
Learning from similar orders in the past to quantify a potential built-in momentum is the first step. The next one would be to use a trade cost model curve to build a utility function for finding an optimal execution strategy which will balance the momentum and the impact. Ensuring that the trade cost model scales well for various order sizes, aggressive execution strategies and the markets where you trade is essential; otherwise the suggested strategy might be far from optimal. A robust volume model is also important to assess the duration of each potential execution strategy and thus the exposure to the expected momentum.
The optimal strategy will be defined by the target participation rate and might be characterised by potential improvement (profit) comparing to historical results, probability of profit and other statistical characteristics to ensure there are no skews and abnormalities which could result in a large loss or other undesirable results.
It is worth emphasising that the improvement and the probability figures should be assessed in the context of the entire order flow, and not on any single order. It is a numbers “game” and the more you play it, the higher is your chance to win.
To focus the effort, the highest return on investment is to identify not only behaviour patterns of portfolio managers, but to focus on cases where the trader and the portfolio manager are not aligned. Thus, if a portfolio manager has behaviour patterns and the trader knows them and adjusts the execution strategy accordingly, you might be able to reinforce the trader’s appropriate behaviour and potentially highlight short-term biases to the portfolio manager.
However, if the trader is not fully aligned with the portfolio manager, our research, based on over 200 portfolio managers, based in various countries and continents, shows that there is a 5-25 bps range for potential improvement (profit) on 10-20% of the order flow from most PMs.
Let us take all orders for a hypothetical portfolio manager over the last few years. If you look at the distribution of daily stock momentum (side adjusted) from order arrival time till 5 days later, you can see that it ranges widely between – 3% to +3% per day (Fig.1).
Executions for this portfolio manager can’t be always passive or aggressive, due to the wide range of built-in momentum.
If you look at duration and participation histograms of the executions for this PM (Figs. 2 & 3), you will notice that most orders have been executed over a day and some over 2 days. The duration spike is around 6 hours, which suggests day VWAP orders. As a result the participation rate ranges widely with most orders executed passively.
Historical execution slippage versus arrival was 26 bps +/- 104 standard deviation, while interval VWAP versus arrival was 26 bps as well +/- 105 bps. It suggests that the firm is most likely “VWAPing” its orders. When you apply PM profiling you can identify the orders where history contains a signal that is statistically significant.
The signal strength can be measured as potential improvement in bps juxtaposed against probability of positive results (Fig. 4). The higher an order scores in both dimensions, the stronger the signal.
If you use only those orders to learn from history, then the results from your PM profiling reduce the slippage to 12 bps and the standard deviation is down to 95 bps (Fig. 5).
This result is achieved for about 20% of the trading value ($1.1bn out of $5.6bn) and the overall improvement is about 10% ($1.5m out of $15.8m) (Fig.6).
When you look at the profit opportunity cumulatively you want to see it consistently growing over time (Fig.7). It’s easy to see that the Executed (dark blue) strategy closely matches Interval VWAP (light blue), which confirms the day VWAP assumption about the execution style of this trading desk. When you apply PM Profiler (dark green) you can see that over time it has been consistently outperforming an actual, executed strategy.
You can further enhance the profiling strategy, by limiting multi-day executions not to exceed 3 days. This is illustrated by the Adjusted profiler strategy (light green line). It is easy to see that this adjusted strategy keeps most alpha since the light green line is very close to the dark green one. This adjusted strategy also ensures that the profiler is close to existing trading practices and the risk profile of the asset manager. Limiting the maximum number of execution days also reduces standard deviation from 104 bps to 95 bps, as you can see in Fig.5.
If you apply the profiling across all PMs at the firm, thus multiplying potential improvement by the trading volume of the firm (no high frequency firms are in this sample), it could translate to an opportunity to increase their returns for clients by tens of millions of dollars, euros and pounds annually.
Based on the research published by Bloomberg over the years, 83% of trader success comes from reading the momentum correctly. Therefore PM profiling should be at the top of the list for decision support systems for the modern trader.
Portfolio managers can also be interested in understanding their short-term momentum patterns, so this tool can and should serve both the trader and the portfolio manager.
Broker-Algo profile
Once you have identified the optimal execution speed, you can analyse historical executions to see which Broker-Algo combinations have been the most successful for an order type you are about to trade.
To help you profile Broker-Algos, some trading platforms provide peer universes containing millions of routes contributed by hundreds of buyside firms in a non-attributable way over the last few years. When a firm decides to contribute its history, it benefits in return from the entire community experience.
The same concept of multi-dimensional analysis and supervised machine learning could be used here. Profiling of algos can be slightly different comparing to the portfolio managers, though. For instance, most algos are sector and side agnostic – two important factors for PM profiling. Spread, however, and target participation percentage are important factors for an algo selection.
The country might play a role in algo selection, since brokers adjust their algos across regions and countries. Thus an algo from a broker can carry the same name across markets but behave differently and yield different results.
As for measuring Broker-Algo results, optimal algo selection could be based not only on historical performance against a chosen benchmark or a set of benchmarks, but could also take into account consistency of historical performance and the number of observations. In some cases the outcome could be a group of Broker-Algo combinations that have been yielding similar results.
An important nuance to consider is that the same algo can behave very differently depending on the parameters selected. A passive algo, which reaches and invokes an ‘I would’ criterion, can become very aggressive and thus drastically change the algo’s behaviour. Mixing up these aggressive executions with passive ones will create a huge noise and an inconsistency in algo analysis.
Thus, you have to take parameters’ selection into account to understand the algo’s behaviour and to reduce the noise in historical observations. This is key for a successful Broker-Algo profiling, which can, according to industry advertised results, lead to 2-3 bps of outperformance.
Venue analysis
After you have identified an optimal execution speed and the best Broker-Algo combinations for a particular order, you can afford to drill down into these algos to see what they are doing under the hood.
You should quantify the efficiency of anti-gaming mechanisms, protection against adverse price selection and front-running. Ensuring there is no conflict of interest between the best execution obligation and the collection of maker/taker rebates is part of this analysis.
If you choose to customise Broker-Algos, you will be able to compare your results versus standard algos to measure the improvement. Custom algos might limit the ability to use peer universes and will narrow the analysis to the client specific routes only.
Once you have an optimal speed and the best algo, you should expect a potential improvement left in the venue analysis to lie in the execution tactics, and not the execution strategy. You should expect a smaller profit from venue analysis compared to PM and algo profiling.
Learn from the past
Based on the above-described research and assumptions, multi-dimensional order profiling which aligns the portfolio manager, the trader and the market can generate the highest returns. This should be the top decision support tool for the modern trader.
Broker-Algo profiling can ensure the best match between your order characteristics and selected strategy. It has the second largest potential impact. Venue analysis could be a beneficial tactical tool to complement the two previous ones.
The three aforementioned decision support tools can work separately, but using them together will bring the most powerful results in maximising trading profit.
Using big data and advanced decision support tools can allow the trader to be at the vantage point and have better control in executing his order. Portfolio profitability will naturally improve, so why not learn from the past to make better pre-trade decisions?
As William Faulkner famously wrote: “The past is never dead. It’s not even past.”
[divider_line]©BestExecution 2015
[divider_to_top]
