The Gateway to Algorithmic and Automated Trading

Anatomy of an Algo: Beer Necessities

Published in Automated Trader Magazine Issue 27 Q4 2012

When an unexpected news event lands right in the middle of a high ADV trade, most execution algorithms struggle to adapt strategy to the sudden change of circumstances. But as Brian Schwieger, head of EMEA Execution Services Sales and James Wardle, Vice President, Execution Services at Bank of America Merrill Lynch explain in this example trade from the brewing industry, there are exceptions to every rule.

Brian Schwieger

Brian Schwieger

James Wardle

James Wardle

The Scenario: It is August 7, 2012, a day that will resonate with shareholders of Heineken NV for some time to come, as the takeover of Asia Pacific Breweries (APB) takes a new twist. Heineken has enjoyed premium success in recent years; it now stands as the third largest global brewer with a strong geographically diversified franchise, a rare gem in many portfolios.

A portfolio manager of a large conventional asset manager in London took the view in early October 2011 that Heineken's diversified franchise could be a significant factor in driving market outperformance. While developed markets were likely to prove flat, those in Africa and Eastern Europe could still support overall sales growth. Over the period of several days, she therefore acquired a holding of 1 million shares at an average price of EUR 33.12.

Since then the company has seen a steady flow of good news; mid December 2011 saw the company's purchase of the Galaxy Pub Estate from RBS. This was followed by further upbeat news, with annual results announced in mid February 2012 beating forecasts with a 9 percent increase in annual profits to EUR 1.58 billion.

Finally, the recently approved APB acquisition would help cement Heineken's emerging market presence and has helped contribute to the stock's price approaching a five-year-high. After this strong run, the portfolio manager feels the time is right to take some profits. She therefore instructs her firm's trading desk to sell 320,000 shares with a limit of 44 EUR.

The Asset: Heineken NV ("Heineken")

The Challenge: To sell 320,000 shares of Heineken over the next trading session with a limit of 44 EUR minimising implementation-shortfall cost. The order is large at approximately 20 percent ADV and with the backdrop of recent activity in this name, excessive volatility is likely to be present. The key to execution success here will revolve around spotting the right signals at the right time.

The Algo: A liquidity-seeking implementation-shortfall driven algorithm which employs a variety of opportunistic and protective signals to source favourably priced liquidity intelligently from all sources in the market-place (exchanges, dark pools, MTFs and internal crossing), whilst dynamically balancing trading-speed with market-impact and alpha-decay.

Signals used are selected based upon trade difficulty and market conditions and look to exploit the autoregressive/autocorrelation relationships across time-series components to anticipate future volume/volatility movements, as well as using intra/cross-asset class correlations for trend and fair-price prediction, in addition to looking at the impact of news sentiment and arrival rates. An important differentiator is that the algorithm focuses on opportunity rather than a fixed trade schedule.

Upon receipt of an order, the algorithm checks volume and stock statistics (intra day and historical data including liquidity, spreads, market activity and volume information). Then - based upon a combination of the urgency level specified (see below), order data and stock data - it calculates a unique trading solution specific to the order and time of day. It then reviews this constantly throughout the trading session, adjusting according to market conditions.

Key Parameters:

• Urgency: A value from 1 to 5 offering a range of execution speeds and styles, spanning from (1) high discretion with very passive lit market activity (typically 5-10 percent of volume) and full dark exposure, through to (5) aggressive liquidity seeking order-type (50-70 percent of volume in lit markets) leveraging advanced order-book tactics

• Get done price: A price beyond which the algo should aggressively sweep available liquidity until order completion or price condition expires

The Trader: The trader heads the asset manager's London trading desk and in view of the relatively large percentage of ADV that the order represents, opts to handle the order himself. While the asset management firm deals with several brokers, the trader feels that the characteristics of the order are best suited to this particular execution algorithm.

Trade Schedule:

07:30 (London time) - The trader enters the order with an initial urgency level of 2 (around 15 percent of volume). The broker's execution desk acknowledges the order. The algo assesses trade difficulty, estimated price impact, and market conditions to plan the initial schedule.

07:55 - The algorithm has been carefully placing orders into the opening auction to maximise queue position and liquidity capture, whilst remaining sensitive to theoretical volume and price information.

08:00 - The stock opens down at 45.285 (-43bps) with low uncross volume (4,364). The algo executes 673 shares in the market and internalises a further 209.

08:40 - Heineken has been slowly ticking up, dragged by market beta. The algorithm has detected this trend and has accordingly adjusted order-book tactics to work the near side and capture multiple spread levels. It executes a further 23,000 shares (including a natural cross of 6,000).

08:50 - Past the wide spreads of the morning and with favourable liquidity picking up, the broker's sales-trader recommends an increase to an urgency level of 3 (around 25 percent of lit volume), with which the trader concurs.

09:05 - Positive momentum persists (now up 40bps from open) and the algo has increased trading speed to benefit (~30 percent of lit volume) looking to capture liquidity in parallel across lit venues and non-displayed blocks in the dark protected by the broker's fair-value price model. Cumulative number of shares executed: 46,454.

09:30 - Broader rally now backing off with speculation on upcoming macro announcements. The algo reduces its participation rate after detecting a stock reversal signal. The order-book is relatively thin and signals are kicking in to optimise microstructure forecasts. The algorithm executes a further 35,671 shares.

10:55 - Prices bounce back from lows as anticipated by the reversal signal and are now trading in a tight range with support around 45.25 and resistance around 45.34. The algo seizes the opportunity to take advantage of the correlated trade price movements to increase liquidation at the top range.

12:00 - Algo is on track to complete with just under 50 percent done and performance looking good. Cumulative volume executed: 154,246.

12:11 -

• Fresh news hits the tape - a rival bid for APB has been launched by Thai Group at S$55 (vs. the S$50 Heineken bid)

• The algo news signal detects a drop in short-term sentiment on Heineken and anticipates a market shock with high price impact and a significant volume increase with gradual decay

• Alerts are flagged to the broker's execution desk and client

• The algo begins aggressively sweeping available liquidity from lit and dark venues and re-calibrates its limit order placement model to optimise orders for an impending drop

Graph 1: Market Volumes and Prices (Source: BofAML, August 2012)

Graph 1: Market Volumes and Prices (Source: BofAML, August 2012)

12:16 -

• The stock is now down 2 percent amid significantly higher volumes, but with the news sentiment signal kicking in early, the algo has been able to unload significant volume at good levels before many other market participants even noticed the event

• Large price move triggers a further alert to the client trading desk, where the head trader decides to amend the urgency level to 1 in order to look for natural non-displayed blocks with low lit participation and ride out the volatility

• Cumulative volume executed: 267,509

12:40 - Volume is beginning to subside and there is a respite with prices clawing back 1 percent from the bottom (-3.5 percent).

The algo is responding to the changing volatility profile, optimising slicing and pricing for the uncertain market regime, while the stock oscillates in and out of limit. The algo executes a further 35k shares, mostly from natural crosses.

12:58 - Final fills are obtained and the ticket completes, executing the full 320,000 shares at an average price of 45.04. The broker's execution desk confirms the result to the asset manager's head trader.

Graph 2: Algo Cumulative Fill and Average Time Step Interval Participation Rate (Source: BofAML, August 2012)

Graph 2: Algo Cumulative Fill and Average Time Step Interval Participation Rate (Source: BofAML, August 2012)