Dark Pools and Price Discovery

July 18, 2014

Block trades have enormous potential to reduce transaction costs, but only a fool would display a large order in a lit market. Meanwhile, price discovery doesn’t happen in the dark, and the probability of a block trade occurring in a dark pool is pretty close to zero. Fortunately, much analytical work will be devoted to determining pricing methods that encompass the risks and rewards of both displaying and aggressing orders.


Here’s the riddle of institutional equities trading: Experience and reason suggest block trades have enormous potential to reduce transaction costs. However, block trades don’t happen in the dark. And only a fool would display a large order in a lit market. So block trading falls short of its potential.


But there’s an answer to this riddle. It’s called price discovery.


For a period of time, algorithmic trading provided an attractive alternative to other methods. Over several years the market lacked tools to interpret and exploit the trail of market data left by large institutional algorithms. But that era is over and has been for some time.


To be sure, block trades also leave a trail of market data. But large trades are inherently more difficult for third parties to interpret, as there are fewer of them. Plus, with large trades, more of your order is done before the market has a chance to figure you out.


Back to the riddle. The probability of a block trade occurring in the dark is pretty close to zero. Yes, there are examples – great ones – in which traders received large fills in dark pools. But those aren’t really trades. They are coincidences. Two particles colliding. In fact, in the stock market, that’s all that ever happens in the dark – particles collide. When the particles are small and there are lots of them, the probability of a coincidence is high. When the particles are large and few, the probability of coincidence is low.


That was the lesson of Optimark, the great noble failure that took the equity trading world by storm 15 years ago. The brainchild of the electronic trading pioneer and successful innovator Bill Lupien, Optimark intended to solve the riddle by revolutionizing the price discovery process. It offered the promise of a market-wide optimization of supply, demand, and price. All market participants would submit something called a satisfaction density profile – your degree of happiness if you were to trade across a range of prices and sizes, up to your maximum.


Optimark had standard profiles for different types of traders. A market-maker profile would be downward-sloping – satisfied buying larger quantities only at lower prices. An investor’s profile might be the reverse: willing to pay higher prices, but only for more size. Optimark received everybody’s profile and arrived at the prices and quantities that maximized satisfaction across the market. At the time I wondered if traders were really interested in maximizing the mutual satisfaction of all traders, as opposed to just maximizing their own.


Not unlike IEX today, Optimark was all anyone talked about for a while. The publicity was massive, and expectations ran the gamut. In the end, a number of factors made it difficult for Optimark to achieve critical mass. The product was pretty complicated. It faced formidable political opposition from the New York Stock Exchange. Optimark’s expenses were high, and it couldn’t sustain its burn rate long enough to break through.


The big Optimark takeaway is that price discovery doesn’t happen in the dark. To produce more than mere coincidence – to produce a trade that adds value and reduces cost – one side of the trade has to make itself visible before the trade. That’s the lesson: One side has to go first and make itself known. One trader has to take a risk, to stick his neck out. The other side aggresses. That’s how a trade takes place. One trader offers a choice. Another trader chooses – paying the spread and trading price for immediacy.


Not that Optimark had a hard time getting people to go first. A lot of firms participated over a sustained period of time. The lesson of Optimark is not that it’s hard to get someone to submit orders; the lesson of Optimark is that nobody aggresses in the dark.


This is not as obvious as it may appear. After all, why not submit super-aggressive dark orders? If you’re working an order with alpha and willing to pay up to trade in large size, just make each order immediate-or-cancel with a large minimum fill. Either the trade gets done or no one knows you tried. Where’s the harm in that?


The obstacle is psychological, but it’s real. Traders expect aggressive orders will be filled. If a trader fires the harpoon a few times, he expects to hit something. Come away with nothing and, after a while, he gets tired of trying. Nothing more complicated than that happened with Optimark. In the dark there’s no difference between a near miss and a far one. Aggressors got tired of the misses. And they moved on.


So block trades reduce transaction costs, but they require light, as traders only aggress against block orders they can see. But light can trigger signaling if you don’t trade and can attract adverse selection if you do. Isn’t that an intractable problem? Not necessarily.


For instance, a 1991 paper on “Sunshine Trading” by Stanford professors Paul Pfleiderer and Anat Admati argued that benefits may accrue to institutional traders and their counterparties when they “pre-announce” the intention to trade. Their assumption was that pre-announced merchandise was perceived to be “informationless” – that no one would go public pre-trade with the orders with the most alpha. But in the real world, even that isn’t clear.


In today’s market structure there are two precautions you can take to display block merchandise safely: one, limit your display to a high-value universe. Reg ATS provides the regulatory framework to allow technology to be deployed that addresses this need. An alternative trading system can qualify its participants on a need-to-know basis and only display to the most likely potential counterparties. Two, use price both to mute the signal and to compensate for the cost of signaling and adverse selection. A low-ball bid does not signal much urgency. Yet that bid can provide an urgent counterparty with an opportunity to exchange price for time.


Yes, every once in a while an urgent counterparty is going to run you over, and inevitably those will be the trades that the portfolio manager notices. But they will be balanced by trades that go your way, and the net bias against you will be built into the spread you charge. Consider that both parties are going to trade anyway. Imagine the contribution if, in these situations, you could capture 15 basis points of negative slippage on even just 10% of the order.



This is where future innovation is likely to take place. In the years ahead much analytical work will be devoted to determining pricing methods that encompass the risks and rewards of both displaying and aggressing. These methods will be incorporated directly into algorithms and will be deployed on the desktops of traders executing trades by mouse or by phone. In the process, trading will become more efficient and costs will fall for both parties to the trade. That’s the benefit of leaving a thinner trail of market data for third parties to analyze.


In short, dissatisfaction with algo trading as it exists today is making much of Bill Lupien’s vision relevant again. You heard it here first.