The Gateway to Algorithmic and Automated Trading

Merger Arb Flip

Published in Automated Trader Magazine Issue 28 Q1 2013

One certainty of markets is uncertainty, so even established investment strategies don't necessarily play as expected. Gary Stone, Chief Strategy Officer, and Yurij Baranksy, R&D Head of Cross-Asset Arbitrage Trading and Options Trading at Bloomberg Tradebook illustrate an example from the world of merger arb, where a manager's new twist on a trade creates a major pairs challenge for his trading desk.

Gary Stone

Gary Stone

Yurij Baranksy

Yurij Baranksy

The Scenario:It is the evening of Monday September 3, 2012 and a Paris based hedge fund manager is considering the effect of recent events on his merger arb position in Xstrata/Glencore. In common with a number of other alternative asset managers, after the bid was announced in February he did not immediately take the conventional merger arb position of shorting the acquirer and going long the target, but inverted it instead. His reasoning was that at 2.8 shares, the initial approach was unlikely to succeed and that spread between the stocks was likely to expand rather than contract until the Glencore offer was improved. An additional, more technical, consideration was that Glencore has extremely limited float and that the cost of borrowing stock for a short sale was much higher than the norm - approaching 10%.

Another fundamental factor in the manager's thinking was that Xstrata was likely to remain weak as it was in a state of transition between the closure of exhausted mine assets and revenue from new assets (such as its Antapaccay project in Peru) coming online. Combining these considerations with his reservations about the spread converging until a better offer was tabled, the manager instead opted to reverse the usual convention and went long Glencore and short Xstrata in a ratio of 2.8:1, completing the purchase of 1.4m Glencore and the short sale of 0.5m Xstrata by early May at an average spread of 11.821.

At the close of trading on Monday September 3, the spread had reached 139.62, so the position was highly profitable. However, events of the previous week have inclined him to consider completely reversing this trade and replacing it with a larger conventional merger arb position of short Glencore/long Xstrata. The previous Tuesday Norway's NBIM (a sovereign wealth fund with a nearly 3% stake in Xstrata) made clear that it regarded the current terms as insufficient. On Thursday, Qatar Holding (Xstrata's second-largest shareholder) followed suit by reiterating its original June objections to the deal and confirming that it would vote against the proposed scheme at the meeting set for September 7. Although Glencore's CEO has insisted that he will not increase his firm's offer, the fund manager feels that an improved offer is highly likely. The pace at which other major shareholders are building positions could prove a major obstacle to any attempt to postpone the deal and reactivate it at a later date. He therefore takes the view that an improved and ultimately successful bid will be forthcoming, possibly shortly before the shareholders' meeting set for Friday.

At today's closing prices, the trade has already captured a gain of 127.8 on the widening of the spread since the position opened. The manager thinks that this could continue to expand from its current level of 139.62, possibly as wide as 175 in the short term, so to completely invert the spread position now would both sell to close and short sell to open into a market running in his favour.

The manager therefore instructs his head of trading to close the existing position during the following day's trading session, Tuesday September 4 2012. If at the point when the closure is completed the spread is still below 175, then the trading desk is to also continue selling Glencore and buying Xstrata to establish a net position of short 1.75m Glencore and long 0.625m Xstrata. However, if the spread is above 175 upon exit of the original trade, then the trading desk is to take no further action, as the manager will instead look for an opportunity in a later trading session to short the spread at a higher level.

The Assets: Glencore and Xstrata Ordinary shares

The Challenge:To sell 3.15m Glencore and buy 1.125m Xstrata during the trading session of Tuesday September 4, 2012.

The Algo:The execution mechanism is actually a dedicated pairs trading tool that automatically accesses the appropriate execution algorithm(s) for each leg of a trade. Its core concept is a 'show' leg which is the security (typically the least liquid or more difficult to execute) that is posted in the market. This is linked to an 'anchor' leg, which is not posted but whose price determines the price at which the 'show' leg is posted in the market. The user can specify which is the 'show' leg (if both securities are very liquid it is possible to show both legs), or can let the tool automatically determine the most appropriate choice.

The default setting is for aggressive pricing, so if for example the 'anchor' leg is being bought, its asking price and available volume are used to determine the posted 'show' leg price and size. The intent of this is that if the 'show' leg achieves a partial fill, then there will be a readily marketable (usually limit) order for 'anchor' leg trade that has a very high probability of being filled.

However, other price aggression settings are available. Even if the spread is immediately marketable, users can opt instead to join the bid/offer and wait until they are lifted/hit on one of the legs. A further intelligent aggression option automatically scans the trades executed in the previous 60 seconds to determine whether the market is moving towards or away from the trade. If the former, then the tool will post liquidity to pick up price improvement. If the latter, it will be more aggressive and seek to lift/hit bids/offers in order to complete the trade.

In order to minimise legging risk the tool includes a maximum hedge risk parameter, with which the user can constrain the maximum notional amount or the maximum number of shares that the two legs can be out of balance (unhedged) during trade execution.

Another leg risk feature is the ability to handle situations where, just as a partial fill is achieved on the 'show' leg, the market price for the 'anchor' security suddenly moves away. The user can specify several preferred ways of handling this contingency:

• Wait a specified number of seconds to see if the 'anchor' leg price comes back the outstanding order is filled at the correct price.

• If it is still not filled, the price is made more aggressive for a further specified period of seconds.

• If it is not filled after the second waiting period, the user can either do nothing or execute the leg via a market order. (If the user opts to do nothing, they can manually execute a rebalance after the rest of the order is complete.)

In addition to price aggression, the tool also offers finessed size aggression. If a user has specified a maximum hedge risk of 500 shares, but the market is posting 300 shares, then the tool will take that liquidity anyway (rather than waiting for 500 shares to be available) in order to keep the overall order moving. If the user is well acquainted with the security and knows there is significant hidden liquidity available, they can opt to disable this functionality and insist that the tool exactly complies with their specified maximum hedge risk value.

Merger/takeover deals often have triggers that, if tripped, change the nature of the offer from being (for example) all cash to stock plus cash. The tool therefore allows the user to set price parameters that accommodate these, so that if one of them is tripped both legs of the trade are paused. A similar 'trip and pause' functionality is also available to handle sudden sharp price movements in favour of a leg that would otherwise fundamentally alter the characteristics of the overall trade.

The maximum hedge risk parameter also indirectly controls the type of algorithm used. If a small amount (say 100 shares) is specified then the pairs tool will automatically just use intelligent limit orders. However, if the amount is more than twice the average trade size in a particular name then the trades are routed via a smart execution algo which can access dark as well as lit markets in order to improve price.

The Trader:The trader allocated the trade by the firm's head of trading specialises in merger arb pairs transactions, which are a major element in the fund's activities. He typically layers in/out of positions by placing large slices of the total trade into the pairs tool and replenishing them as they are completed.

Trade Schedule:

(All times are UK)

7.30am: The trader opts to put 10% of the total trade size into the pairs tool initially and will replenish this with similar slices as required. He also specifies a substantial maximum hedge risk size of GBP10,000 notional value, so that the trades will be handled by the smart execution algorithm. The access to dark execution this will deliver is particularly useful in the case of Xstrata which in the preceding week has seen an increasing level of dark executions, now exceeding 5% of lit volumes traded. He also wants to minimise execution constraints, because he needs to allow enough time to open a new short position in the spread, in case it is below the manager's level of 175 when the current long position is closed.

On the basis of history he would normally specify Glencore as the 'show' leg, but opts to let the pairs tool decide. This detects that while long term daily Xstrata volume in cash terms comfortably exceeds Glencore, in the past week the positions have actually reversed and Glencore is currently more liquid in cash value terms. He therefore accepts its suggestion of making Xstrata the 'show' leg.

8.01am: Good flush of liquidity on the open on the LSE, where the execution algo sells 35,500 Glencore and buys 5,500 Xstrata. Algo also sells 7,000 Glencore and buys 1,100 Xstrata on Chi-X CXE. Glencore trades completed at an average price of 387.02, Xstrata trades at an average of 948.4.

8.13am: The spread on the existing position has been rallying strongly over the past ten minutes and the algo has sold 5,000 Glencore and bought 1,200 Xstrata.

8.22am: The pairs tool's decision to make Xstrata the 'show' leg has been vindicated over the past nine minutes, with very patchy trading and liquidity in the stock. Despite this, the algo manages to buy a total of 10,125 and 330 Xstrata on lit and dark venues respectively. Corresponding disposal figures for Glencore are 85,750 and 10,125. Glencore has been rallying strongly, so the average sale price for these blocks was a respectable 388.85.

8.29am: Significant selling pressure on decent volume has emerged in the last few minutes and Glencore has started to come off. Xstrata is still chopping to and fro on spotty price action. The trader believes that Glencore may start slipping back towards where it started the day, and as the 'anchor' stock may be vulnerable to partial fills. He is concerned that this may disrupt the overall trade and therefore decides to change the pairs tool's wait parameters to wait two seconds, increase aggression, wait two further seconds, do nothing. He doesn't anticipate major weakness in Glencore and so is happy to let the trade get out of hedge, expecting to rebalance with a manual sell later in the session when the stock (hopefully) strengthens.

8.57am: The trader's earlier view is confirmed.

Glencore and the trade spread have both reverted to near the day's opening levels, while Xstrata is virtually unchanged. Glencore started to move sharply lower about 10 minutes back, but prior to that the algo had managed to sell a further 171,000 shares across a mixture of lit and dark venues. However, a further 112,500 Xstrata shares have been acquired since 8.29am, so the trade is now significantly out of hedge with a total of 307,250 Glencore sold and 130,755 Xstrata bought, representing a ratio of approximately 2.35 to 1, instead of 2.8 to 1.

9.22am: 83,550 Glencore sold and 22,100 Xstrata purchased in the last five minutes. Xstrata has mostly being travelling sideways but Glencore has started to rally strongly on reasonable volume, so the trader is optimistic that there may be an opportunity shortly to rebalance the position manually. He also resets the tool's wait parameters so it will execute a market order after the two pauses rather than doing nothing.

Figure 1: Glencore 1 minute bar chart for Sep 4 2012

Figure 1: Glencore 1 minute bar chart for Sep 4 2012

10.10am: After a strong rally on solid volume Glencore is now range bound at around the 388 level. The trader's hunch was correct - in addition to 240,500 Glencore sold by the algo since the last update he has managed to manually sell a further 64,050 and has 'journalled' this trade into the pairs tool, so the hedge equilibrium has been restored. 95,500 Xstrata bought.

10.21am: Sharp fall in Xstrata has allowed the algo to pick up a further 12,500 shares at an average price of 940.55 with nearly a third of this in the dark. A further 35,000 Glencore sold. The exit of the original long Glencore/short Xstrata position is now just over half completed.

11.58am: Lit market activity in Glencore has picked up over the past 90 minutes and especially in the last half an hour. The algo has also disposed of several substantial blocks of stock, resulting in total sales of 585,050 since the last update. Xstrata purchases for the same period total 208,600.

12.10pm: Improving liquidity in both stocks, but especially Xstrata, has allowed the pairs tool to complete the exit of the original long Glencore/short Xstrata position. As the spread is currently trading at 143.96 (well below the manager's 'take no further action' level of 175) the trader starts to enter the reverse short Glencore/long Xstrata position. A total of 1,400,150 Glencore sold and 500,300 Xstrata purchased since the start of the trading session.

12.44pm: The spread is continuing to expand and the algo continues to sell into this with a further 220,550 Glencore sold and 78,550 Xstrata purchased. With less than four hours to go before the close of the trading session and the trade barely half completed, the trader sharply increases the maximum hedge risk size to GBP50,000 to help maintain execution momentum.

1.36pm: After rallying strongly for nearly an hour, the spread has eased off to around the 158 level. Over the same period Xstrata liquidity has increased dramatically with 839,669 shares traded. However, despite the more relaxed parameters, because Glencore liquidity does not follow suit, the execution is still rather constrained. The algorithm could probably have traded 125,000 Xstrata during this period, but this would have required executing approximately 337,000 Glencore - the equivalent of nearly 40% of the LSE liquidity at the time. Nevertheless, the algorithm manages to sell 220,000 Glencore and buy 75,000 Xstrata.

Figure 2: Xstrata 1 minute bar chart for Sep 4 2012

Figure 2: Xstrata 1 minute bar chart for Sep 4 2012

2.47pm: Substantial increase in Glencore volume since last update, with well over three million shares trading on lit venues. 365,000 Glencore sold, 134,100 Xstrata purchased. Spread has mostly remained range bound between 158 and 160, with average price achieved 159.27.

3.13pm: A sharp fall in Xstrata's price has popped the spread to 169.46 - the highest level seen since the initial announcement of the bid/merger in February. Fortunately, this has been accompanied by a similar bulge in liquidity - especially for Xstrata. 348,100 Glencore sold, 124,320 Xstrata bought.

3.52pm: Aided by a strong rally in Xstrata, the spread has fallen back and is trading to and fro around the 160 level. Liquidity in both stocks has remained good on both lit and dark venues and the trade completes with more than thirty minutes to spare before the close with the sale of 596,200 Glencore and the purchase of 212,730 Xstrata.