Countless reams of paper and miles of type have been expended on the profound effects of algorithms on market structure and trading styles. The focus predominantly has been on broker market share of algorithmic flow and trying to make sense of the seemingly endless array of new bells and whistles (primarily aimed at the buy side) that brokers and vendors constantly are churning out.

But how profitable has the algorithmic adventure been for brokers? Has investment in hiring quantitative mathematicians and building out distribution and sales networks been worth it? Or have algorithms become such a common feature in the trading landscape that it simply is unthinkable for any broker worth its salt not to offer them (and eat the cost) because that's what clients want? And what effects has the rise of algorithms had on brokers' internal organization?

Not surprisingly, brokers are not particularly forthcoming when asked these questions directly. But it is possible to draw some reasonable conclusions from the information that is available.

What seems certain is that no broker can be taken seriously today unless it at least offers the basic algorithms - volume weighted average price (VWAP), time weighted average price (TWAP), implementation shortfall and arrival price. It's also clear that algorithms are more cost-effective for low-maintenance trades, and that has meant head-count shifts and reductions on sales desks. But the decision to offer a more complex and comprehensive suite of strategies is a much bigger dilemma.

"Anyone that does want to offer a comprehensive brokerage solution has to offer algorithms at some point," contends Harrell Smith, director of the securities and investments practice at Celent. "Whether or not the investment has justified the cost savings is one question. Another, bigger question is what is the opportunity cost of not getting more business, of maintaining future and current market share in a slim-margin and fairly commoditized business?"

But Is There Demand?

Smith's question becomes especially prescient when the common wisdom is that most buy-side customers are not interested in, or sophisticated enough to understand, more-complicated offerings. While clients have developed a "tin ear" for algorithms as they have been constantly assaulted with new sales pitches from their brokers, Smith says, as with investments in securities, time in the market still can trump market timing - if a firm is willing to pull previously developed strategies out of the mothballs once the market is ready.

"We had a strategy called ResRisk in the late '90s, a strategy that was about large list trades and recording tracking error against an index," recalls Tony Huck, head of electronic trading at Investment Technology Group (ITG). "Nobody was ready for it, beyond a handful of clients. Now, there is a much larger audience. Over time, the buy side catches up. When you are writing algorithms, you have to ask yourself, are you just creating to create because somebody thinks it's neat, or are you solving clients' problems? We think keeping it about 20 percent in the ivory tower is a good balance."

Until the broader market is "ready," some brokers' significant investments in higher-level math will continue to be based on the supposition that a few sophisticated clients, such as hedge funds, will do a lot of algo trading. For other brokers, algorithms simply are another way of drawing in business, and even though they are loss leaders in and of themselves, they ultimately help the bottom line.

"For the small, midtier and niche brokers, it is fine and a good business decision to go with vendor solutions, charge a flat fee and embed some key algorithms in their system," Celent's Smith says. "But it's a perception thing - if you are Morgan Stanley or Goldman Sachs, you cannot just be out there with VWAP and TWAP; you have to say you have seven to 12 of these things and they are all being constantly refined." Officials at Bank of America, Deutsche Bank, Sanford C. Bernstein, Morgan Stanley, Goldman Sachs and Credit Suisse declined comment for this article.

There's no question that brokerages are seeing more and more order flow come in via algorithms. Both JPMorgan and EdgeTrade officials say that about 40 percent of their order flow is derived from algorithms. Deciphering how much of a brokers' profits come from algorithms, however, is challenging, if not impossible. Still, a quick, unscientific survey of recent broker financial reports indicates that many of the large firms are doing something right, and algorithms surely are part of the equation.

Lehman Brothers reported record net revenues of $3.1 billion in its capital markets division in the second quarter of 2006, up 38 percent from the same period in 2005. Goldman Sachs' equities trading turned in $2.5 billion, up from $1.11 billion a year ago, a 152 percent increase, against a 22 percent rise in commissions. Credit Suisse's trading revenues were up 52 percent, and Morgan Stanley achieved sales and trading net revenue of $4 billion, up 76 percent from the same quarter in 2005.

Adding Internal Value

It's important to remember algorithms typically enable large-scale brokers to derive as much value from internal use as they do from outside business, points out Carl Carrie, VP of new product development at JPMorgan Securities. In recent months, JPMorgan has been trading 90 million shares per day through internal use of algorithms, about triple the flow that comes in directly from clients on peak days, according to Carrie. "About 40 percent of all of our order flow is executed algorithmically. We are now amortizing the benefits from increased efficiency, best execution and improved efficiency," he says. "That is usually left out of the discussion. To think that algorithmic trading is just the anonymous FIX flow business would be a misunderstanding of the economic advantages."

To make investments in algorithms worthwhile, the algorithms must be fully utilized across the business, Carrie stresses. The client-flow business has low margins and high-fixed costs - including hardware, market and client connectivity, data center build-outs and trade cost analysis, among other expenses. Algorithms aid trading firms greatly in statistical arbitrage trading, event arbitrage on block desks and hedging against risk, Carrie relates. Algorithms have been so deeply embedded in JPMorgan's business for the last eight years - long before they were offered to clients - that it is virtually impossible to quantify their impact, he adds. "The benefits of algorithmic trading go far beyond any one business," Carrie says. "It is a style of trading and not a separate business."

Further, bulge-bracket brokers with proprietary trading desks can use statistics based on internal algorithmic flows to determine transaction costs and market-impact costs, Carrie continues. They then can pass this information to their clients. He acknowledges the concern buy-side clients have about their information being used against them. "Electronic trading is an alternative way of paying for other services," Carrie says. "We still live in a bundled world. But the measures the sell side takes to guard against those issues are pretty extensive."

The Agency Advantage?

Firms that run on a strict agency basis - such as Instinet, EdgeTrade, NYFIX and ITG - believe that one of the main attractions of their businesses is that their nonproprietary stance means algorithms serve the customer alone. "Our feeling is that over time, algorithmic trading is something only agency brokers will be in a position to deliver to the marketplace," says Joe Wald, CEO of EdgeTrade. "Firms that conduct proprietary trading that have algorithmic trading do not perform as well."

It is difficult to evaluate the success of agency brokers versus trading firms attached to massive investment banks, where soft-dollar and bundling arrangements drive algorithm use and the embedding of algorithmic trading into general trading results obscures algorithmic data. Still, there are indications that, irrespective of the results algorithms get for clients, agency firms are performing respectably.

For example, ITG's commissions rose 31.4 percent to $117 million in Q1 2006 and net income doubled. Instinet, since the absorption of the INET ECN into Nasdaq, is no longer a public company; however, the company selectively reported financial and volume figures in Q1 2006 showing a rise of 20 percent in trading volume to 133 million shares.

While agency brokers may be gaining momentum, the established full-service brokers still dominate the algorithm market. Aite Group estimated that agency brokers held about 30 percent of algorithmic trading market share in April 2005.

Meanwhile, algorithm innovation continues to offer returns for firms with the scale to absorb the costs and reap the benefits. There are indications that clients do respond to the bells and whistles - if they are well-suited to demand. More than 50 percent of JPMorgan's electronic flow, for example, is derived from its Trading Algorithmic Optimizer (TAO), released in November 2005, which manages trading at both the portfolio and single-stock levels, the company's Carrie notes.

Turf Wars

However, the road to the state of Zen implied by the name TAO has not been entirely smooth in the brokerage business. The costs incurred by trading firms in algorithmic development have been human as well as financial. As was the case when program trading first arrived on the scene 10 years ago, turf wars have broken out among traditional phone-based sales traders on the cash equities desk and the shiny new algorithmic trading units, according to a former managing director of a trading firm.

"Now, you are seeing program desk and algorithmic desks blend together - and you are still seeing turf wars there," the executive says. "There are a lot of neat stories flowing around about how buy-siders have three people from the same firm calling them, sometimes trashing one another. [The sales people] don't want to be marginalized - I blame management. A lot of firms have tried to tell their people, 'You need to sell [algorithms],' but they don't give them an incentive."

JPMorgan's Carrie acknowledges that many firms have struggled with the transition, suggesting that it must be well-managed to be successful. "There was some natural cannibalism of traditional order flow by electronic flow during the transition," he says. "That is happening less and less. The electronic flow business is increasingly accepted as a natural course. It is now becoming more accepted for high-touch transactions."

Now that head count has stabilized and algorithms have become integrated into brokers' business models, often algorithmic trading is happening without clients' direct involvement - at human-trader prices, executives say. "More and more these days, the sales trader the buy side is trading with, paying 3 to 5 cents a share, he turns around and inputs it into an internal algorithm," says the former managing director. "So clients are trading more with them whether they know it or not."

Whatever opinion the buy side may have of revelations such as these, it seems clear that algorithms are firmly a part of the brokerage business, and brokers are always finding new ways to make them more profitable. And as long as clients receive reasonably good executions, they seem content to send other business to brokers, even as they demand more transparency into execution quality and bundled services. If these trends continue, the role of algorithms at brokers may expand - regardless of whether brokers can draw a direct correlation between algorithms and their bottom lines.