Decimalization has decimated displayed liquidity. Large bids and offers have virtually disappeared. Traders are too worried about getting pennied' to post size. Now many frustrated pros are clamoring for Washington to reverse course and institute a five-cent minimum trading increment.
But even while pros are calling for nickel ticks, most large trading desks are quietly adapting to the new reality. Strategies are emerging that help traders cloak their intentions and avoid unwelcome market impact.
Underpinning many of the new methodologies is the practice of breaking up large orders into tiny bits and pieces. The trend is underscored by the numbers: Since 2000, the average trade size on the New York Stock Exchange has fallen from about 1,200 shares to about 515 today. On the electronic Nasdaq market, the trend is less pronounced. The average trade size has declined from about 680 shares in 2000 to 610.
Slicing and dicing is not new, but the level of sophistication by which it is accomplished has grown. Using powerful computers and complex algorithms traders can feed huge orders into the market almost undetected, they say.
At the forefront of this algorithmic trading trend is Credit Suisse First Boston. The bulge bracket player first began testing various "order staging" strategies on its proprietary desk in early 2001, the year decimalization began. Later that year, CSFB expanded the use of its algorithms to customer orders. In 2002, its Advanced Execution Services (AES) group was born. It offered select clients direct access to its algorithms from their desktops.
Dan Mathisson is a CSFB managing director and head of AES. The exec joined the CSFB proprietary desk as a statistical arbitrage trader in 2000 after spending eight years with quant shop D.E. Shaw. Mathisson spent a year on CSFB's prop desk before transferring to the program desk and developing AES.
Mathisson has been involved with algorithmic trading in one form or the other for the past 11 years. He was interviewed by Traders Magazine Technology Editor Peter Chapman.
Traders Magazine: How did AES evolve? Why do this? Mathisson: We decided to roll it out because of decimalization. We saw a huge change in the way that the markets function and in the way people trade. The number of prints went up by about six times from what it had been in the fractional world. The average size showing went down by a factor of more than ten. In the fractional world, a typical market was, say, three-quarters to seven-eighths, 25,000 [shares] by 50,000. Today, it's 75 cents to 78 cents, 300 by 800. We've seen quoted size drop tremendously while the spread has gotten narrower. So, it has become a very difficult market to trade. You can't trade in big pieces the way you used to. Within this organization there was a need for a way of breaking up the orders and cutting them into little pieces. AES is an attempt to automate the trading process. To make it more efficient. And to cope with decimalization.
Traders: O.K. Mathisson: So, around that time we were building a system on the prop side that did a lot of these things and automated parts of the trading process. We had done a lot of work on order placement. We had done a lot of analysis as to where within the spread to place an order. What kind of opportunity costs you get in different types of trades. What is the proper way to measure impact cost. At the same time there was this need to somehow get a tool into the hands of traders and salespeople and clients. To cope with decimalization. So, toward the end of 2001, we began converting our proprietary algorithms for client use.
Traders: The problem was in how to place or stage an order? Mathisson: Right. We had to change the way we placed the orders. In the fractional world, you could put out an order of say 10,000 shares. In a fairly liquid stock, 10,000 shares had very little effect on the marketplace. But in the decimal world, even if you put out an order that used to be considered small, say, 3,000 shares, a tremendous percentage of the time you will get pennied. You will be showing a 3,000-share bid for 75 cents. And someone will bid 76 cents for 300 shares. Just a penny above you.
Traders: And you won't get the stock? Mathisson: Well…You get jumped. People jump in front of your order all the time. You used to be able to just pick a price and put it out there. Orders that used to be considered relatively small are difficult to trade now. Our solution is to cut the orders up into tiny pieces so they hide in the flow. But when you do that you can't just spray them out mechanically. Or you will get killed. You will get gamed. You will get picked off.
Traders: O.K. Mathisson: We spent a tremendous amount of time trying to figure out the right way to put out these little pieces. When should you put them out? What price limits should you put on them? When should you cancel them? How many should you have out at one time? In other words, how do you put these orders out in a way that you'll end up with a good execution?
Traders: If you just shot out a fixed number of shares, say, every ten seconds, someone on the floor would figure out what you were doing? Mathisson: Right. Let's say you had something mechanical; some kind of mechanical time-slicing strategy. And every minute I send out 500 shares at the market. The guys on the floor are smart. It won't take them long to figure out that every minute they'll get another 500. Other players can pick up on it besides the specialists on the New York. OTC market makers will pick up on it. Block traders. Stat arb traders. If you are doing something obvious, people will pick up on it. And they will game you. And you will lose money. And get a bad execution. We spent a tremendous amount of time trying to figure out how to make the system behave in a way that is unpredictable and not gameable.
Traders: Specifically, AES is a group of seven different trading services? Mathisson: No. It's really one comprehensive trading service. It's organized by trading tactics. Each tactic represents either a trading goal or a trading benchmark or a trading technique. On top of that, the user can layer constraints. He can layer things like minimum and maximum percent of volume. Start time and end time. Aggressiveness. Price limits.
Traders: Can you give me an example of a typical order? Mathisson: A user can put a pretty complex order into the system. Buy 70,000 shares of Ford. Be at least five percent of the volume. But be no more than 20 percent of the volume. Start at 11:30. Be done by 3:00. Put a $17 top on the price. These are pretty complicated orders. So, I wouldn't call it seven trading services. It's seven different methodologies that the client can choose from.
Traders: VWAP, price improvement, reduction of market impact…A trader can choose his strategy? Mathisson: That's right. Traders have different benchmarks for different orders. Some orders they want to spread out over time. Sometimes they want to minimize implementation shortfall. Sometimes they just want to beat a particular limit price. Sometimes they want to beat the closing price. Traders have many different motivations for trading and many different reasons for choosing [a particular strategy]. Each methodology is tied to a benchmark. And the trader gives us his benchmark.
Traders: Traders can add their own instructions? Mathisson: They can layer their constraints if they have any. But the fewer constraints the better the average performance will be. Constraints can only hurt you. If there was some constraint that was positive we'd already have that. You are taking maximum advantage of the algorithms we've built when it's in its unconstrained format.
Traders: Do you see more single stocks or baskets? Mathisson: Most clients are sending us a series of single stocks, but we do receive baskets as well. Trading single stocks is frustrating due to decimalization. That's what I hear over and over again from clients. That prior to decimalization they wouldn't have even considered using this kind of product.
Traders: Now they have problems. Mathisson: Since decimalization they have been really struggling to cut the orders up small enough. They're faced with a choice: Do I sit there and piece it out every couple of minutes myself and type a million orders down on the keyboard? Or do I throw out an order that's too large for the market, get pennied and take the loss? Here we are providing them with something in between. They can fire the order out once and then forget about it. But at the same time the order will get worked in lots of little pieces. Traders: What kind of customers are you getting? Mathisson: It's a very diversified client base – hedge funds, mutual funds, pension funds, other broker dealers…
Traders: Are you trading more listed or Nasdaq securities? Mathisson: It's a mix. It's interesting though. We find we do slightly better – when we control for variables like average daily volume, order size, market capitalization, volatility – in OTCs. Across the board. Not a lot better. About one tenth of a cent better. That tenth of a cent is sort of a tax imposed by the listed marketplace. By the structure of the marketplace.
Traders: Whereas in the OTC? Mathisson: In the OTC, there is nobody in the middle of the transaction. It's the ultimate fair marketplace. Since the Order Handling Rules and the advent of electronic trading, the OTC market has become incredibly fair. It's similar to the E-minis, the futures. The OTC compared to the listed marketplace is similar in many ways to the E-mini futures compared with the traditional contract traded in the big pit.
Traders: By the humans? Mathisson: Right. In the E-minis, if there is an offer and you hit the button a tenth of a second ahead of some other guy, you get the fill. It's the same with the OTC market now. If I go to take an offer in Arca…If our system shoots out that order one-tenth of a second ahead of some guy we get the fill.
Traders: But with the listed market? Mathisson: With the listed marketplace, that's not necessarily the case. If one order is a tenth of a second ahead of the other one… Both orders arrive, but it's still possible neither of them gets the fill. The fill could go to someone in the crowd, for instance. Whatever the reason, the structure of the listed marketplace seems to impose a cost of about a tenth of a cent.
Traders: Nasdaq used to be a big middleman market. Mathisson: But Nasdaq is a very pure marketplace nowadays. Very fair. You have the five major ECNs – Arca, Brut, B-Trade, Instinet, Island. Plus SuperMontage. It's very easy to stitch it all together. We use an internal system called Pathfinder to aggregate all the ECNs. When you are looking through Pathfinder, it looks like one clean marketplace. And it is. If the stock is offered at 14 cents and the machine tries to take it… Nearly 100 percent of the time it will get done on at least the size that was showing.
Traders: But not on the listed? Mathisson: There we get done only 92 percent of the time. So, eight percent of the time [the quote] was there and we tried to take it, but it just mysteriously fades. Of course, some percentage of that time somebody got ahead of us.
Traders: So, it's 50-50 split? Mathisson: We do do a little more on the listed side. Clients come to us because they want us to solve a trading problem. And clients have more trading problems on the listed side nowadays than they do on the OTC side.
Traders: The New York's Direct+ seems made for you. Mathisson: It's still slow. You can only come in once every 30 seconds. There are all these rules that they built around it. But whenever the New York comes out with a new product we always examine it. Experiment with it. If we find it will improve our trading performance then we implement it. But I can't comment on specific tools.
Traders: Do you provide data to clients who want to check their performance? Mathisson: As often as the client would like. Every order ever sent to us by every client is saved in a database with its performance stats. We can sort and filter those any way the client chooses. If a client wants to see how he did in his Nasdaq trades of between 10,000 and 30,000 shares over the previous three months. If he wants to know his performance vis-a-vis his arrival price. It's just a couple of keystrokes to pull it up.
Traders: Do you participate in any of these trades as principal? Or is it pure agency? Mathisson: AES is an agency product. Our business model is very straightforward. Very clean. We provide the client with a great execution and demonstrate that with performance stats. We try to demonstrate to the client, statistically, that we are adding value. Clients pay a commission.
Traders: How much? Mathisson: I can't give you a hard number. But the commission is less than that of a traditional full-service broker. CSFB is a full-service brokerage, of course. But this is an automated product that involves no human interaction. So, there are efficiencies on both sides. For the buyside and also on our side.
Traders: It's all do-it-yourself? They don't call you? Mathisson: Direct access orders typically do not involve phone calls. But we have a telephone hotline for support and clients can always call their traditional salesperson.
Traders: You test the algorithms internally before you roll them out to clients? Mathisson: Yes. It's an evolutionary process. In order to move our algorithms forward we bring a lot of in-house flow in through the system. The very first thing we did before we gave it to a single client was to put it on the desktops of every block trader and every sales trader. We put it on the desk of every single person who trades U.S. stocks within CSFB. That's over 400 internal users. They pounded on it and gave us a lot of criticism when they found a situation where the algorithms were not performing the way they should. We would get suggestions from traders as to how they felt it should work in the situation. We generated a whole pool of ideas.
Traders: Your traders are still using it? Mathisson: Sales traders, private client guys, block traders, convertible traders, derivatives traders, international traders, program traders, index arb traders. Every trader has access. So, we get suggestions from these people that we are looking to test. We then take the internal flow and run 10 percent of it through an experimental tactic. Each tactic has both a production version and an experimental version. We constantly look to see if it is doing better or worse. Once we have a statistically significant positive number then we start putting more flow into it. Over time the strategies that are empirically succeeding, doing well in the marketplace, get more and more of the internal flow. The strategies that are doing worse, get less. Once we've established that something is actually better we then roll it out into production for the clients.
Traders: So, testing never stops. Mathisson: We're refining all the time. Algorithms are never finished. Algorithms can always be better. Until the day comes when the machine is buying everything on the low and selling everything on the high, we'll be trying to make them better.
