Complex market conventions plus multiple pricing factors plus low volumes; three factors that add together to give you - no algo. That, in short, is what you'll find if you're looking for execution algos for illiquid commodities.
Those for more liquid commodities - for example, energy commodities like crude and natural gas - have been around a while. But it's only relatively recently that commodities markets have become fully open to electronic trading - a prerequisite for algo trading - and, perhaps oddly, the attraction of the rarities hasn't triggered the evolution of correspondingly 'inquisitive' algorithms.
Auto may be a prerequisite for algo, but it is not in itself sufficient. Try finding traders already using execution algos for illiquid commodities and you're likely draw a blank. Lack of appetite? Possibly. But more probable is the sheer diversity of variables they need to factor in. These, at least, are becoming relatively clear for the pioneer in the field.
Compared with those for illiquid securities, execution algos for illiquid commodities simply have to do more. For one thing, they need to measure both volumes and volatility: more volatile (because less liquid) commodities require models that build signals to protect against losses.
"...if there were still trading pits, they would be full of armchairs"
"We haven't had much demand for something like that," says John Hyde, execution consultant at Bank of America Merrill Lynch - though he points out that the bank's Ambush algo "discreetly probes the market and executes trades when urgency is high and impact is of paramount concern".
Some of this complexity comes from the fact that behind these commodities are physical assets. [Even the terms 'liquid' and 'illiquid', as applied to commodities, are of questionable accuracy, according to Luke Jemmett, who is responsible for the market in energy-related commodities, including carbon and wet and dry freight, for interdealer broker GFI. Jemmett points to a relatively liquid German power market compared with emerging Eastern European markets that are significantly less liquid.]
Either way, these are not bonds, as David Allen, commodities sales and trading manager at AVM LP, puts it. A natural gas operation, say, is one component of an interconnected system linked to multiple factors and prices across regions - the price of transportation, insurance and spot fuel, weather-implied demand, the forward price on the curve, the demands of power plant and retail customers.
"Traders need to take into account cash versus physical settlement, delivery location, grade, expiry date and the rest," says Allen. "In the case of the US, if the delivery location of the gas is some point in Louisiana rather than in southern California, you have different geographies, infrastructure, weather and price. In contrast, in the UK, gas is delivered and indexed to a national balancing point in a virtual location. The devil is always in the detail. Market convention is often very important.
"Most trading operations aren't comfortable modelling these systems and building such an algo in-house because it involves too many complex and distinct variables," he adds. "These systems are not fungible with other financial non-tangible systems."
The lack of fungibility has created another problem in the development of execution algos for illiquid commodities, suggests Allen. Pure tech developers tend to see it as an adjunct to existing asset classes, rather than a qualitatively distinct algo.
"Algorithms are hard to write," he says. "The danger is that, if left unchecked, technology purists might treat execution algos for commodities like any other code, with the naive assumption that an asset is an asset; that if you can track it, chart it, and data aggregate it, you can trade it; and that market conventions are roughly the same for any market."
What they need instead, he claims, is an appreciation for the scheduling and operational detail, decision support tools and time series programmes necessary for training and support.
"Investing in commodities for me could mean acquiring physical power plants - buying, operating, restructuring and, potentially, later on selling the physical asset. The problem is that most people active in the market come from a financial background, not a true infrastructure background. If you were to do a rough count, you'd find that 90% of commodity algo traders had absolutely no background in physical commodities."
Even if it's not possible to develop a commodity-execution algo by tweaking, say, a fixed income one, say, Allen points to cross-asset systems that can identify and monetise differentials between interrelated commodities, where triangulation may exist with equity and FX markets.
"As beta becomes more expensive and individual commodity markets resume exhibiting their own unique price and volatility profiles, quant traders will be forced to explore higher order implied spreads between assets," says Allen.
One advantage of bringing algos in-house is that they're tailored to the needs of (potentially multiple) markets, according to Günter Tschiderer, a fund manager with Sigma Commodities, a BNP Paribas Investment Partners commodity business. "I'm very open to algo execution. Why not?" he says, though he adds that the house does not allow him to trade in illiquid commodities.
In any case, execution algos have their work cut out. The technology needs to be capable of modelling "down to a tick", says Ben Jackson, senior executive vice-president of Sungard's energy and commodities business. "In terms of execution platforms, speed is critical, as is the ability to mitigate post-trade risk. The challenge is to understand risk in multiple dimensions, such as position, margin, P&L and market risk, and understand this instantly."
Speed is one issue; alerting the market is another. These are trades that keep their heads down.
One problem algos for all illiquid assets share is the potential for market impact. This makes many forms of more aggressive algos designed for the more liquid equities and exchange-traded futures markets unsuitable. At the same time, argues Richard Bentley, general manager for capital markets at Progress Software, more passive algorithms such as electronic eyes ('snipers'), which take smaller quantities and then back off and wait for the market to repopulate, can be used effectively in illiquid markets.
GTI provides windows when customers can come in and place a price without others seeing it. In theory, this will encourage participants to place orders on the open market rather than trading bilaterally or via dark pools. It's only when the trade is executed that it becomes apparent to the market.
The problem is timing risk: the longer the period you execute over, the more potential for the market to move away from you. In any case, illiquid markets require a pre-trade impact analysis to understand how the trade will move the market.
Ray Murphy, chief risk officer at Nautical Capital, believes that all this leads to continued reliance on human intervention. Placing and cancelling trades as you try to gain an edge on the wide bid/ask spreads indicated on-screen results in "a lot of phantom liquidity", which can play havoc with algo trading systems.
"That is part of the reason I have always tried to manually handle at least portions of the trading," says Murphy. "I can be more specific in both the price and timing of execution." He adds that most of the algorithmic trading done at Nautical is based on end-of-day pricing. The availability of trade settlement orders makes these types of algorithmic systems much easier to execute, he claims.
GFI traditionally used a hybrid model that combines technology with voice. If you use an electronic system, there may be spread between bid and offer, says Jemmett. "Technology is good because it aids transparency and liquidity but there are things you can't do. Depending on the market, the split between electronic and voice trades is slightly different."
In the meantime, to bring illiquid commodities even close to the securities market, you need arbitrage, says Douglas Hepworth, research director of Gresham Investment Management. The problem is that "tangible" (that is, illiquid) commodity markets are not arbitrage markets.
If you have futures versus physical assets or versus other related futures, electronic trading makes it easier to capture small discrepancies. If the futures are cheap, you buy the futures and sell the component stocks; when they're rich, pari passu, you sell the futures and buy the stocks.
"This does not work in the tangible commodities markets," Hepworth points out. "If you see discrepancies in cattle versus cattle futures, you could buy cattle and sell futures, but with much greater cost and more complexity. And on the other side, you can't short live cattle. That closes off the normal arbitrage flow; it is at best an asymmetric arbitrage."
A potential solution to the problem of illiquidity is to build liquidity. This sounds tautological but it acknowledges the point made by Jackson - that the technology itself isn't enough: optimal execution depends as much on the modelling maturity of the algo trader.
At the simplest strategic level, there are arbitrage opportunities from a single exchange, with spread contracts versus futures and winter/summer contracts for natural gas.
A second model comprises execution across multiple exchanges, involving multiple data sources - for example, for spreads versus futures on fungible contracts that trade on NYMEX and ICE. In this case, the technology and data requirements introduce another level of complexity.
An additional model seeks arbitrage opportunities with exchange-traded funds (ETFs) versus underlying futures, such as GLD versus Gold Futures. These involve more complex risk modelling because they involve futures contracts on multiple exchanges.
The final model involves markets with less liquidity. "We're seeing some of our customers looking at these markets," says Jackson. "If you look at heating oil contracts, the liquidity hasn't been there. So there has been starting and stopping of strategies here. It's unclear whether we'll see the volumes but we're definitely seeing moves from our customers, as well as exchanges pushing customers into this area."
Algo trading introduces liquidity into these markets. "What you have in every market is that it's difficult to be first - to put your strategy against bringing that kind of liquidity into the market. If you take emissions, you have listings on multiple exchanges, which presents arbitrage opportunities. But you haven't seen sufficient volumes or volatility yet."
If not now …
There are some important distinctions when it comes to execution algos for illiquid commodities. The first is between traders who would like to use them and those who couldn't care less. It's a pretty even split.
It seems reasonable to believe that there will eventually be an execution algo for pretty much every tradable asset class. But illiquid commodities are close to the bottom of the list. One buy-side trader identified "at least five" algos he wanted to see developed before illiquid commodities - including Treasury futures and long bonds.
"Equities will always have more client demand but developing algos for futures rounds out the offering," says Hyde. "We looked at the trend and saw the open-outcry model going by the wayside."
Yet demand will probably increase for a commodities algo with a boosted investor appetite for the asset class. Jackson points to a growth in algo trading within the context of the growth in futures markets compared to a decline in equities, pointing out that the CME recently posted year-on-year gains of 90% and ICE, 12%.
Moreover, some of the most illiquid commodities have been the ones to recover most quickly, from a price perspective, after 2008. The speediest recoveries came with some of the most esoteric: palladium, vegetable oils, rubber and other softs - the ones Allen describes as "very unattractive commodities from a media standpoint - the equivalent of the redheaded stepchildren of oil and gold".