In Hyde Park Global's arbitrage strategies, for example, the spreads move dynamically, according to Afshar. An adaptive model, he says, will measure the spreads in terms of a time series: It's not just the dollar amount that matters, Afshar notes; how long it takes for the spread to close also is an important factor. "Sometimes the right spread could be $2 and sometimes the right spread can be $10," he says.
Currently Hyde Park Global executes trades on NYSE Arca, BATS Exchange, EDGX, NYSE and INET (Nasdaq), Afshar reports. "We don't deal with dark pools," he adds, explaining that dark pools do not execute quickly enough for the firm's arbitrage strategies and also charge higher commissions than the public markets. "It's more expensive to do dark pools, and for us transaction costs is a pivotal aspect of our profitability model," Afshar says, noting that exchanges and ECNs provide rebates for adding liquidity. "All of our models rely on our high-speed trading system to find available liquidity and minimize market impact to keep total transaction costs as low as possible."
Arbitrage and Co-Integration
According to Afshar, Hyde Park Global conducts two variations on arbitrage strategies. The first involves taking advantage of cross-exchange mispricings on multiple venues. "You have to be very fast -- these arbitrage opportunities don't last very long, and they're small," Afshar says, stressing that the trades must be small so as not to move the price of the stocks.
The firm's second approach to statistical arbitrage involves the relationship between what Afshar calls "co-integrated stocks." He points to the example of a man walking a dog to explain co-integrated stocks: "Imagine I'm walking my dog in the park -- you see me and you don't see my dog, but you know he is near me," Afshar says. "It's impossible for you to predict the movement of the dog or where I'm going in the park. But you know there is a relationship between me and the dog and that we will go home together."
A similar relationship could exist between stocks such as Lowes and Home Depot, for example. "You can trade this relationship," Afshar comments. Based on historical data, "You have a sense of the range of the distance between the stocks' [prices]," he continues. "The proprietary trading firm uses computers to identify what range is the optimum and what range of spreads is outside of its known patterns."
But with extensive amounts of data that needs to be processed, of course, the computers must be able to draw from databases. "Any adaptive model using large amounts of historical data needs a database, and databases introduce some level of latency," Afshar acknowledges. "The models are dynamic and they need to adapt -- I don't think there is any way to do it without the database."
To mitigate the latency introduced by its large databases, Hyde Park Global taps complex event processing to detect patterns in streaming data. According to Afshar, the CEP technology minimizes the latency within the firm's statistical arbitrage models because it performs calculations in memory rather than pulling from a database. Then, "We turn around and store it, and it's stored in a way that doesn't create significant latency," he explains, referring to proprietary database storage and retrieval methods.
Despite the firm's sophisticated adaptive models, Afshar concedes, statistics can never predict extreme situations such as the credit crisis because there is not enough data to predict these events. "No statistical model is going to show you that because there are just not enough crashes [to provide data] to make these events statistically predictable," he notes.
Nonetheless, regardless of what happens in the markets, humans are never allowed to intervene with Hyde Park Global's trading platforms, says Afshar, who is the only person with such authorization. And the only time Afshar would intervene, he insists, is if there's an extended power interruption in the office. "We don't second guess the program," says Afshar.