The Gateway to Algorithmic and Automated Trading

High pressure performance

Published in Automated Trader Magazine Issue 15 Q4 2009

One of the mantras on any trading desk these days is “do more with less”. Sabrina Rovelli, US Equity Product Manager at Bloomberg Tradebook, explains how predictive modelling can accomplish this by taking execution algorithms to the next level in terms of productivity and performance.

Professional execution desk traders pride themselves on leveraging their specialist expertise to deliver alpha when handling stock acquisitions/disposals. In a perfect world, conditions would enable them to maximise this. Their workload would be perfectly allocated and balanced so that they would always be handling a manageable number of orders in stocks where they had a strong understanding of the relevant technical and fundamental drivers.

Reality, however, paints a different picture. In practice, traders often find themselves having to cope with unexpected surges in the number of orders; often orders are in stocks that are unfamiliar. Under these conditions, a conventional execution algorithm is limited, as it typically cannot deliver the predictive intelligence of an experienced human trader. As such, it is similar to a highly featured and sophisticated car that lacks a driver to anticipate road conditions and hazards.

Assistant, not replacement

In view of that, the logical next step is to add a measure of predictive intelligence capable of steering execution algorithms in a manner that assists, rather than competes, with human traders. When execution algorithms first appeared, many traders saw them as a threat to their job security. Over time this perception has changed; algos are seen as useful productivity tools; specifically, algos are viewed as an "extra set of tools" in the execution tool-kit that can be actively managed to deliver superior execution performance. Therefore, any new tool that seeks to add predictive modelling to algorithmic execution needs to perform in a similar manner, by acting as the trader's assistant - not replacement.

In order to achieve that, the predictive capabilities need to be well-integrated with existing execution algos and intelligent order routing. If they aren't, then the potential workflow benefits will be lost and execution quality may suffer as the trader will be obliged to manually adjust the algo and routing settings based on the guidance of the predictive model.

But how good is your algo?

While accurate prediction is valuable in its own right, execution performance ultimately depends on the quality of the algo it interacts with. Having a good idea where the stock will be in five minutes time is of limited use if the underlying algo is dumb. To seek the best results, especially in the U.S. equity markets where liquidity is spread across lit and dark market venues, the underlying algo has to offer smart posting and intelligent order routing.

If the algo is closely monitoring real time trading and is capable of quickly shifting orders to venues showing the highest liquidity, then it can take full advantage of the intelligent prediction. For example, in the previously mentioned situation, the predictor anticipated that the market was poised to break down; it therefore made sense to grab as much liquidity as possible while the opportunity was there. The algo needs to maximise representation while seeking to minimise information leakage through techniques such as hidden orders and randomised anonymous acronym display orders, in order to maximize the predictive advantage.

Think like a trader

Any effective price predictor needs to function in much the same way as an expert human trader. It will therefore take into account a variety of factors, such as market depth, consolidated tape activity, current activity on the trader's own order, bid/offer spread/size, technical indicators and any imminent reports/earnings releases. A further factor, crucial when attempting to predict short term price behaviour, is relative performance - i.e. how is the stock performing relative to its sector and the market as a whole?

Figure 1

Figure 1

Source: CQG, inc. © 2009. All rights reserved worldwide.

Such performance has a tendency to mean revert. For example, Figure 1 shows a 5 minute GE bar chart with below it the moving linear regression of the difference between GE's and the Dow's percentage returns, together with the difference's long term average. As can be seen, while the stock frequently over/underperforms the index, this does not persist for extended periods, nor diverge completely.

Bloomberg Tradebook's B-SmartSM AUTO

An example of the type of integrated intelligent assistant outlined here is Bloomberg Tradebook's B-SmartSM AUTO. This solution provides advanced trader support with a tightly-integrated statistical price prediction model to Tradebook's proven B-Smart execution algorithm. B-Smart already offers four different aggression levels (REACT, PASSIVE, NORMAL and AGGRESSIVE) that the trader can specify according to his/her view on current conditions in a stock. B-Smart AUTO moves things to the next level by enabling the aggression settings to be directly controlled by the price prediction model. For example, if a trader is working a large buy order and the predictor anticipates the stock is about to break down sharply, it will automatically become more passive, move from normal to react mode, and wait for the market to fall before re-engaging execution.

This characteristic is particularly valuable for fine tuning switches between execution modes on the core algorithm. If the stock is strongly outperforming the market and the algorithm is in aggressive mode in order to maintain the necessary acquisition rate, sooner or later the stock will exceed a significant threshold. (For example, two standard deviations above the long term mean of relative stock versus sector/index performance.) When it does, this might influence the predictor model to switch the algorithm's execution mode back to passive, in anticipation of the stock pulling back and range trading.

One size doesn't fit all

A robust price predictor can obviously add huge value to execution performance and desk workflow. However, it cannot forecast individual circumstances, such as the urgency of each trade. These circumstances would obviously affect the way in which a human trader would handle a trade, particularly the points at which he/she would switch algo execution modes.

Therefore, to deliver maximum value, any practical predictor model needs to be able to accept a parameter input from the human trader indicating trade urgency. The predictor can then take this into account along with stock performance to determine the optimum execution mode switching points for the underlying algo.

Putting it all together

Combining statistical price prediction with a smart algorithm is highly significant in its own right. However, it is possible to further maximise the benefits by making the new technology part of a multilayered execution strategy. If the trader's EMS is capable of managing cascading "What if?" scenarios, then assorted price/volume/trade completion/news triggers can be set that will flip between strategies when triggered.
For example, if a stock price is above a maximum threshold, the order could be handled by a combination of predictive model and smart algo. Below that level and down to a minimum threshold it could be dealt with by a dynamic VWAP strategy. Finally, below the minimum threshold, the order could be handled by a straightforward arrival price strategy.


The addition of intelligent price prediction to the trader's toolbox is a major step forward. Apart from generally boosting productivity, it offers the sort of multipurpose wrench that can save the situation on those days when reality intrudes that bit too much. Predictors at the cutting edge are also agnostic; they can offer automated price guidance irrespective of a stock's liquidity or trading profile. As a result, they add value across a wide range of trading scenarios - from mid-sized orders in unfamiliar but liquid names, to relatively large (and typically time consuming) orders in thinly traded small caps.


in this document constitutes an offer or a solicitation of an offer to buy or sell any security or
other financial instrument or constitutes any investment advice or recommendation of any security or other financial instrument. BLOOMBERG TRADEBOOK believes that the information herein was obtained from reliable sources but does not guarantee its accuracy.

Communicated by Bloomberg Tradebook Europe Limited, registered in England & Wales No. 3556095, authorized and regulated by the UK Financial Services Authority No. 187492. This communication is directed only at persons who have professional experience in the investments which may be traded over the systems and certain high net worth organizations. Available in all countries and is only offered where clearance been obtained.

Bloomberg Tradebook LLC member of FINRA ( SIPC/NFA. Bloomberg Tradebook do Brasil is the representative of Bloomberg Tradebook LLC in Brazil registered with the BACEN. Bloomberg Tradebook Services LLC, Bloomberg Tradebook Australia PTY LTD ABN 36 091 542 077 ACN 091 542 077, Bloomberg Tradebook Do Brasil LTDA., Bloomberg Tradebook Canada Company Member of CIPF, Bloomberg Tradebook Limited, Bloomberg Tradebook Hong Kong Limited the first ATS authorized by the SFC-AFU 977, Bloomberg Tradebook Japan Limited member of JSDA/JIPF, Bloomberg Trading Services Japan LTD, Bloomberg Tradebook Singapore Pte Ltd Company No. 200104338R, Bloomberg Trading Services (Singapore) Pte Ltd Company No. 200101232G. Bloomberg Tradebook Bermuda LTD, licensed to conduct Investment Business by the Bermuda Monetary Authority.

BLOOMBERG, BLOOMBERG PROFESSIONAL, BLOOMBERG MARKETS, BLOOMBERG NEWS, BLOOMBERG ANYWHERE, BLOOMBERG TRADEBOOK, BLOOMBERG POWERMATCH and BLOOMBERG.COM are trademarks and service marks of Bloomberg Finance L.P.("BFLP"), a Delaware limited partnership, or its subsidiaries. BLOCK HUNTER, BWAP, DEFINING LIQUIDITY, LIQUIDITY DOME and TRIGGER TRADING are trademarks and service marks of Bloomberg L.P. ("BLP"), a Delaware limited partnership. The BLOOMBERG PROFESSIONAL service (the "BPS") is owned and distributed locally by BFLP and its subsidiaries in all jurisdictions other than Argentina, Bermuda, China, India, Japan and Korea (the "BLP Countries"). BFLP is a wholly- owned subsidiary of BLP. BLP provides BFLP with all global marketing and operational support and service for these products and distributes the BPS either directly or through a non-BFLP subsidiary in the BLP Countries. Bloomberg Tradebook is provided by a BLP subsidiary, Bloomberg Tradebook LLC, and its affiliates and is available on the BPS.