Greg Tusar, head of the Goldman Sachs Electronic Trading business in the Americas, sat with Traders Magazine recently to discuss a number of issues affecting equity trading today. Among the topics addressed were developments in market structure like order routing and order handling, as well how algorithms and dark pools are evolving. An industry veteran of 18 years, Tusar offered some of his insights in this first leg of a two-part Q&A.
Traders Magazine: You mentioned that order handling and routing are more complex these days. Could you expand upon that?
Greg Tusar: Now, versus a decade ago, not only would you say that things are a lot different, but there’s no question they’re incredibly more complex, in terms of how orders are handled and routed.
TM: How so?
GT: I think people are struggling with the process that used to be quite understandable, in terms of trader to sales trader, to two-dollar broker or floor broker, to walking out to the post–or the order was handled by a Nasdaq market maker–to now, a process that is highly automated. How exactly that order is handled is not as easily understood by clients.
It goes to a lot of the things in our Concept Release [comment letter], also. We think the way to deal with that is to not necessarily make the structure simpler, because at this point, given how technology has evolved in the market, to try and simplify it is going to be an uphill battle.
TM: Then, what do you do?
GT: The idea is to make it more transparent so people can at least make more informed choices about which brokers to use, based on the information they’ve disclosed–making their own order-handling information transparent back to them, which is something a lot of clients have asked for.
TM: Can you give an example?
GT: A lot of clients have worked with their OMS [order management system] vendors and with us to say: I want to know on every execution where that execution took place. And that is something they’re now able to use–that information–to make more informed decisions about where to go with the next order.
And so, if you’re willing to read the Concept Release response, you might also want to read the Form ATS filings. You know what order types an ATS actually has and how the subscribers use them. This idea that there are dark pools, the more that we can make the exact mechanisms by which those were transparent, the more that will increase confidence in exactly what’s happening there.
TM: Is there a lot of skepticism these days from money managers about where their orders are getting done?
GT: I’m not picking that up. I don’t know if it’s skepticism, so much as it is a desire to understand it better. It’s an interest in really understanding in a world of many choices as there are, and how do you choose among them? It’s a desire for greater understanding. And it also speaks to the difference between ATSs and the exchanges, and so forth.
TM: Give me an example of the need to understand the process better.
GT: So, Rules 605 and 606, which are a little out of date at this point, can be modernized to make the information a little more user-friendly. That might explain to people how, when I put an order into a smart router, exactly how it went about selecting a venue to execute. Or, if it looked at six different venues, which six different venues did it look into? That’s kind of hard if you were to pull up the 605 or 606 reports, today. It doesn’t really tell you that. It tells you at the end of the day where stuff was executed, but it doesn’t tell you the manner in which your order was handled, and how you searched for liquidity. For example, did you ping in six different places before you actually went and removed the offer? Those are the things that clients want to understand better.
TM: Just like the ICI letter said, they want to know where their order went and where it didn’t get executed.
GT: Absolutely.
TM: And they want to know everything. And you’re providing that now, or trying to?
GT: We are absolutely providing that in real time back to clients. One of the challenges, from a technology perspective, is their OMS able to take on the execution … what you can provide back on the execution is the venue of the trade. It’s harder to pass back something like: before you bought stock here, we checked in these six other places.
TM: How does it work at GSET?
GT: A lot of times it’s a dialog with a customer that says: here’s what our procedure is, or here’s how we do it. Or, we provide them with an actual data file that says: here are all the orders that were placed before you got the execution. So those fall under the general rubric for transparency. That’s how we get people more confident and understanding it a little bit better.
TM: But clients still have questions?
GT: Some of the questions we get are: how do you simultaneous manage fees in the maker-taker world and the optimal placement of limit orders? Are they ever in conflict? Or, how does that work? So, what we proposed is passing all of those back to the end client, so that the actual price they experience is the price inclusive of the removal fee or the rebate.
And in the case of a retail order, that would include the payment that end customer receives, which would look like price improvement rather than a payment to the routing firm. What that does is it aligns the end user’s objective, which includes more to do with fulfillment, or: do I care about the acquisition of the rebate, and what’s my real objective? Then, what we’re trying to do and what they’re trying to do is perfectly aligned. That was one of the places where I felt our comment letter was different.
TM: Yes, years ago, when payment for order flow in the retail world mattered, the argument that the discount broker said: when we get this payment, it allows us to charge a cheaper commission.]
GT: Right.
TM: I never really believed that …
GT: It’s hard to really … this would make it very explicit. And then, the thing about being in that regime is it’s all about best execution. It’s about achieving the best price–not separately managing the fees and separately going after … Obviously, it would require a lot of work for the industry to support, but we don’t believe that maker-taker is inherently flawed, but we think this would be a slight twist on it that we think would be good.
TM: Well, you say: it’s a flawed system; that’s what we have. So, what would be the alternative? The only thing you could really do is go back to nickel increments. And I don’t think the industry’s going to do that. Then liquidity providers would come in, because the high-frequency traders wouldn’t be there.
GT: Or, they might. It’s really hard to say.
TM: That’s interesting. They would be there under nickel spreads?
GT: Right. I think that a good place to look at what might happen would be Europe and Asia, where for the longest time they’ve had spreads that were commensurate with stock price. So, they have spreads that are nickels for $20 stocks and dimes for $50 stocks. Personally, I’ve always thought that seems like a reasonable solution.
TM: Can you extrapolate from the buyside learning more and more about how their orders are visiting different destinations to see where that trend is headed? Would the buyside start taking more control of directing where its order flow is going?
GT: I think that on some level, yes. I’ll qualify that. There’s a desire to understand it because there’s intellectual curiosity, and there’s a "I need to trust that my broker’s order routing objectives are in my own best interest." And the only way to do that is to insist on complete transparency with respect to how it works. I’m not sure there’s a complete desire to control it at the micro level. There’s an absolute desire to have it at a level up from that–an algorithmic experience that’s tailored to their unique objectives.
TM: Can you give me an example?
GT: That may be, "I have a portfolio manager with objectives like this. I have another portfolio manager with objectives that look like that. Can you help me construct an algorithm different for this person versus that person, with something that parses out?" I know it’s this PM, so I’m automatically going to route orders a particular way. There’s a lot more of that dialog, I would say, on algorithm customization with the buyside.
Most of the dialog around order handling is educational dialog–and making clear exactly how it works. I don’t think there’s a desire to get down into the weeds and really make changes at that level. It’s more that they need to know that we’re acting in their best interest.
TM: You mentioned the customization aspect. We wrote about a big firm that basically profiles the portfolio manager and, based on that … is that a new trend? Or, is it something you expect to see more of?
GT: I think it’s a trend. It’s something where it’s the next level of customization. The first level was: OK, I understand your algo works this way. This is how I like to trade. Customize the algo to behave the way that somebody thinks about trading. The next level up is: I have an algo selection issue–so, for different types of orders, with different levels of urgencies. Right now I’m manually selecting among eight different choices. How can I automate that a little bit more, so that I’m making more informed choices, or automating it or being smarter about that? That’s a trend a few firms have started, but I expect to see that happening more.
TM: It’s "the other guy has it down the street, so I need it too?" Or, is it that they’re just leaving more money on the table?
GT: No, I think it just helps with algo selection. Right now, there’s a manual process: I think this order has high alpha, therefore I’m going to choose the [right] strategy. It’s making that process a little more scientific. Based on the study’s actual data, I know what the profile of those orders is, and therefore, it’s making it more scientific.
TM: So, would that be the next generation of algorithmic trading development? It’s going more on the data of past trading decisions made by people in the organization?
GT: That’s one of the places. I see others. A lot of that work is a collaboration between us helping do data analysis and the actual buyside trader, some of whom you’ve profiled, working within their own organization. But even within our own algo suite, we still see a lot of things we can do to continually improve. Some of the things we’ve worked on recently have to do with going back to the blocking and tackling of order placement strategies. And you’ll see some research from us shortly that we’ve done on passive limit order placement and the optimal price and location to the placement of passive limit orders.
TM: That sounds right up our alley.
GT: These are things that we find where it’s not something you could check the box on and say "we’re done." It’s a constant process of improvement. So, smart order routing is something we’re continuously investing in. We’re in the process of implementing a next-generation platform, on which both the smart routing and the algorithms are being built that allows us to be a lot more nimble about making client-oriented customizations. It’s our ability to go from the financial engineering of: we have this data set; we worked with the client to profile their orders, to implementing a customized algorithm on our platform should be much shorter than it was with our other platform.
So, we’re probably well into that process. But the point is that I don’t think we’re ever really done, even with the basic algo suite. We’re working on VWAP and squeezing out fractions of basis points of improvement there. We’re working on a shortfall algo to improve on the financial engineering of that. We’re working on the portfolio algo.
TM: How do you improve that–based upon getting new data or seeing anew the way orders are interacting in the marketplace?
GT: Things aren’t static, so where you used to find you’d have one experience of posting orders at the midpoint, you now might have a different experience. So, therefore, you want to be more passive. Where you place limit orders optimally is a function of how the rest of the market’s order routing tables work. So, where you receive the most fulfillment changes dynamically. It’s a constant process.
TM: It’s the quant side of the execution, as opposed to the investing side of it.
GT: I think it’s not something you can build and let stay static. It’s a constantly evolving thing–even VWAP. You wouldn’t think we still find ways to get half of a basis point of price improvement in development of VWAP.
TM: We were talking about full disclosure of where their orders went. The concern was that someone out there would be able to reverse-engineer the algorithms. And that’s why some people don’t want full disclosure.
GT: I suppose so.
TM: How would they know it’s your order?
GT: I wouldn’t worry about that. In the trade-off between: do we give away information about how our algos work or are we more interested in getting clients comfortable with our order-routing practices, we would opt more for the latter than worry about the former.
TM: What’s the next development in order routing, and where are we now?
GT: The next generation, and what you’ll see when we put our paper out, is getting much more scientific about … if there are eight venues and soon there will be 10, how do we choose among them? Is that helping us? If the market is 10 to 12 and the client wants to bid 10, where do you want to place that order? We think there’s a much more scientific way to go about that. In an optimization process that involves looking back historically, going through an actual optimization process–given the client’s urgency and preference, etc.–the next generation is simply taking what we have and being more scientific about it.
TM: It’s almost as if the algorithm and the routing are really the same–there’s nothing new about … a router is a router.
GT: Correct. The next generation is also about making it faster, being able to react in a world that requires microsecond speeds in order to carry out client instructions. And so we’ve done things like, in order to get through all of the options market data–which comes at you at a rate of a million messages a second now–we took however many machines it used to take to consume all that data and worked with a firm to put that in hardware. So, as we’re routing client orders through our options smart router, we’re actually taking all of the market data into the same chip that you’ve got in your Sony PlayStation. That’s also part of the next generation: finding ways to deal with the amount of data in a world in which data rates are only going up. Our next generation will also do that. And it will take the direct feeds and parse them in a piece of hardware, rather than in a piece of software.
TM: So, it’s all about computing power?
GT: Yes. But this is all on behalf of: we have a client order and need to make a decision as quickly as possible on where to route it. Part of it is the financial engineering, and part is the infrastructure work we need to do to make the infrastructure faster to squeeze out latencies.
