The Gateway to Algorithmic and Automated Trading

The Automated Trader Interview

Published in Automated Trader Magazine Issue 24 Q1 2012

There are beginners in this business... and there are beginners. GHF Group is the second kind. Big and global, GHF Group is an established financial services holding company that includes diverse, independent businesses: global clearing, investments, and trading. The trading business includes experienced traders and other market professionals, and has recently started a move into "the business of algorithms" that so far, has looked less like a learning experience for the Group, and more like a potentially big opportunity for the rest of us - for quants, model-builders and just about anybody else with an idea worth automating. Andy Webb went to meet Ron Hertshten, a managing director in GHF Group who is responsible for algo trading, to discuss.

Ron Hertshten

Ron Hertshten: One thing to clarify before we start. In the business of algorithms, we're actually quite new. We have two operating algos that we're working on, and we're testing a few more. We look to this to become a very significant part of our business, but right now, we're quite small.

Andy Webb: Are you talking about algos in the sense of order-execution algos, for finessing order placement, or do you mean alpha-capture algos?

Ron Hertshten: Alpha capture. For order execution, we have quite a few; we call them trading tools. We have a group in India developing those for the traders. They're for internal use. The way we work with the completely automated trading systems is that we have a couple of developers in-house, and we also team up with developers all over, forming joint ventures with them.

Koby Batito and Pavel Ivanchenko

Koby Batito and Pavel Ivanchenko

Andy Webb: On the proprietary trading side, you've got the automated part, which is your department, isn't it?

Ron Hertshten: No! Let me take a step back. I started off recently in the group - I worked here in the past and then returned - I was heading up the office in Shanghai, then I opened another office in Beijing and so was head of our business in China. I am Israeli-American by citizenship. We acquired last year a company in Israel that was involved in proprietary trading. I was asked to come back and head up what we have here in Israel. I'm in charge of Israel and Eastern Europe; we have five offices, three in Israel, one in Hungary, one in Romania. Just recently, we have started developing the automated trading, and that's where slowly most of my focus is shifting. I'll be doing that full-time once it's at a level where it needs that much management.

Andy Webb: At the moment, are you involved in prop trading more generally, or do you have some responsibility for selling clearing services, and so on?

Ron Hertshten: I have no responsibility for clearing at all. I am purely prop trading, and I will be moving over to automated prop trading.

Andy Webb: Just briefly, thinking about the manual traders you've got, are they employees, or on a profit share; how do you work it?

Ron Hertshten: They are full-time employees. We recruit them from the top schools globally. We primarily recruit traders with no experience and then we train them. We don't teach them a specific style, and that's where this becomes really interesting. They bring their own personal touch to the market. They share profits, and that's a huge incentive, but they're full-time employees.

Andy Webb: How did the automated part of this come about? Yours is not a small organisation - where did it come from?

Ron Hertshten and Eyal Ben-Amram

Ron Hertshten and Eyal Ben-Amram

Ron Hertshten: It was a natural progression. It became very clear to us a while back that we needed to have more of an automated element to our trading. There were two ways to go about it. One was to focus on building in-house and one was to go through joint ventures. We decided that the trading tools - execution algos, et cetera - were at a simple level and could be developed in-house, but we thought it would be great if we could team up with very good people all over the globe to form joint ventures for the creation of fully automated solutions. We have two of those that have been developed in-house as well, so we're not neglecting that side of it, but we're looking at the opportunities we can capture because of our global exposure, because we come into contact with great people wherever we are.

Andy Webb: With the stuff you've developed in-house, and with the stuff you're seeking to do as JVs with outside specialists, would you say there's a particular style of automated model that you're after, or are you after a diverse portfolio of approaches and models?

Ron Hertshten: We're not looking for a specific style, but we are looking for something that can definitely be leveraged in terms of size, that we can build on. We're looking less at ultra-low latency, that's not the remit we want to get involved in; we're looking more at the strategy ones. In terms of the strategy, we're open to anything. We come across some very intriguing ones.

Andy Webb: I'm intrigued here. Looking at your organisation and what it does elsewhere, you must have a shedload of clearing memberships. So you must have a physical presence in an awful lot of the co-lo facilities, presumably, or you're at least close to the matching engine. So in many ways you've amortised a lot of the cost that other new entrants to the market would have for high-frequency stuff, because you're not necessarily going to be buying new rack space, new lines, new everything, because you're already there. So I'm therefore doubly intrigued as to why you're not so interested in that low-latency space when you've bypassed a lot of the fixed costs up front.

Ron Hertshten: When we look at it, we know what we are, and we know what we're not. We're a trading company - which is to say, the part where I'm operating is a proprietary trading group inside a financial-services company. But we're not a technology company. We're looking always at what our competitive advantage is. To compete against somebody who's already there, who already has a lot of experience, for us to jump in and compete, the costs are very high; even though we have the rack space, that's only one small part of the cost. We looked at it, and we decided we can bring our competitive advantage to the table in terms of developing the strategies. Obviously, when we develop the strategies, even though we're not looking to be ultra-low latency, the fact that we are that close to the exchange in a lot of locations does give us an advantage in execution. But we're not looking for that to be solely our advantage.

Andy Webb: Again, in view of your having clearing memberships across the globe, do some of your strategies look at fairly sophisticated multi-leg trades that are arbing across multiple markets and possibly time zones, or are they all single-market strategies?

Ron Hertshten: The majority are single-market strategies. We have some that are trading the same algo in more than one market, but not against each other. We have one that does work across markets.

Andy Webb: That's all my preconceptions out of the window straight away. I looked at the organisation, and thought, these guys are in a great place to do A, B and C, and you're not doing any of it. Okay…

Ron Hertshten: We started in '93 as a clearing business on LIFFE. In 2000, when screen trading came in, the big thing was all the prop groups. We were still clearing, and said instead of chasing after clients, why don't we create our own? That's where our thinking comes from. We were still in London, where there was a lot of competition, so we said why don't we do it in Israel? We went in 2002. There, we had the competitive advantage of a lot of really high-quality guys: this was their one shot at the global financial markets and they really gave it their all. In 2004, we moved on to other places. So we started in clearing, now we're proprietary, and we combine the two to create a really unique competitive advantage.

What was interesting, when we started going abroad in 2004, we went to India, China, eastern Europe, places that people associate with outsourcing, particularly the back office. We had our back office in London and all our traders in India. We're used to changing people's preconceptions of how things are done; we're used to thinking a bit creatively.

Itamar Eden, Yael Armon, Ron Hertshten, Musa Mirib and Raphael Swerdlov

Itamar Eden, Yael Armon, Ron Hertshten, Musa Mirib and Raphael Swerdlov

Andy Webb: You have fantastic access to enormous intellectual capital in those countries, as you've obviously found out. With the two models you've developed in-house, where is your alpha development centre? Where do those two models come from?

Ron Hertshten: Specifically they came from Israel, but there's no limitation. We don't have a geographical centre. We're trying to step away from that. We don't want to look at one location because we find culture, creativity, all over - there's no reason why somebody in India, say, can't come up with a great idea to create alpha. Those limits are just limits we place on ourselves; why do that? These two started in Israel because that just happened to be the case, but there's no reason to say that, two or three years from now, Israel will necessarily be the place that created most of the alpha.

Andy Webb: So let me get this clear. Any prop trader, or any person working for the firm in any of your centres, could come up with an idea - could say, this is the approach I take with my manual trading but I think it would automate really well, how about knocking up a model for it? Is that the way it would go?

Ron Hertshten: It could go, yes. We're very happy to explore ideas; obviously we're not going to go after every one, but we encourage our guys to talk to us. They're in the markets, they see stuff a lot of the time that we as managers don't see. We have hundreds of people trading the markets, extremely intelligent and very creative guys, they come up with great ideas. That's one source from which we create algos. The other main source is (because of our global spread we have offices in India, China, Hungary, Romania, Israel, Mauritius, and we have clearing offices in London and Chicago, which gives us a lot of exposure) so we get to meet a lot of people who come to us with algos that are already working. When we sit down with somebody who's not from the group, it's usually somebody who's quite a few levels beyond the idea stage. We're looking to work with them to optimise their model, and bring it to the best assets through our connectivity to the exchanges.

Andy Webb: With the guys in-house, presumably by now people in the organisation are aware that you're the go-to guy, becoming more and more the go-to guy, for the automated stuff. Is the as assumption that wherever they are, you'll get an email? How does it work?

Ron Hertshten and Yael Armon

Ron Hertshten and Yael Armon

Ron Hertshten: We're still hammering out the exact details, but the idea, yes, is that they'll tell their local manager about their idea, that manager will pass it on to me, I'll examine it with my team and we'll think it through from a lot of different perspectives. If we want to do it, we have the option of doing it locally to the guy, adding a quant, say, and somebody to write the algo, or we'll do it centrally.

Andy Webb: With a lot of trading models and trading ideas, there are often a lot of sub-components to the overall model, and you often find that there are bits of a model that are basically re-usable code. It can be applied to other, completely different models. When you were talking earlier about working with your external JV partners, talking about tweaking stuff, are you saying that's your model, let's see if we can help you optimise it? Or do you say, for example, that's your model, I wonder if it would work better if we took this little component that we've used in-house with one of our models, and plugged it into your model? Is it like that, or more hands-off - you're taking their stuff more as it runs, if they've done their proper testing and development, and just doing some final adjustments?

Ron Hertshten: We're looking more at the final adjustments. That's how it is now. It might change. The idea is, we're not looking to build a model from scratch for somebody who comes to us from outside. However, we do have a lot of experience in the market. We have seen things that work and don't work. There are lots of ways that we can optimise models, from IT connectivity through to market access - a guy might have been trading in one market but his model might work better in another, say. We had someone come up to us with an idea: he envisioned it working on the S&P500, we thought it would work better on gold. In terms of the code, that will probably be proprietary to the joint venture; we're not looking to move code from one place to another.

Andy Webb: Yes, that was the next question I was going to ask you. Obviously, there are hugely sensitive areas around intellectual property rights. How do you work that with the JVs? Full disclosure?

Ron Hertshten: Again, we've just started this, so it's a little bit hard to say. I can tell you more about how I envision it happening. In terms of intellectual property, it is a very sensitive area, we realise that. For people we're looking to work with, that's going to be a critical point in their decision-making. We respect that. We don't need the disclosure of the full code. We're looking more to manage a JV from a risk perspective, manage it in terms of how it happens. The code itself - we don't need to see all of it.

Andy Webb: You need to see enough to get your head around what it's trying to do?

Ron Hertshten: Absolutely. We have to understand very clearly what it's doing, what the risks are; we have to make sure that there are certain features in there to handle risk, but we don't have to see everything that it's doing.

Andy Webb: With the stuff that you've developed in-house, which is generally single-market rather than cross-market, would you say that the approaches tend to be based on classical technical-analysis ideas, or more on statistical ideas?

Ron Hertshten: It's more statistical ideas. One, for example, is based on statistics and it's looking at relative prices in the STIR market …

Andy Webb: We're looking at inter-month spreads here?

Ron Hertshten: Exactly.

Andy Webb: That's one. Would you describe that as pretty typical, or do you have straight directional trading?

Ron Hertshten: We have another algo that's purely directional. It's momentum and mean-reverting. It's taking outright positions in a vast array of markets. It depends on the idea and what markets we think are best for that idea. We have a lot of guys trading and they are always coming up with new ideas on how to trade.

Andy Webb: What about the trading platforms on which you deploy the models?

Ron Hertshten: We're using generic platforms for the most part, and we're building on their APIs.

Andy Webb: Like…

Ron Hertshten: TT, Stellar…

Andy Webb: Stellar? That's interesting. I didn't know many people were using Stellar yet. So you're pretty agnostic. It's not like - we always use this platform. Okay. How do you work with your external JV partners? The platform you use, it could be anything?

Ron Hertshten: It could be anything. Obviously we have a preference to work with something we've really known before. But it really depends. If they've already developed something that's really complete and ready to go, we'll find a way to work with that. If they have something that is nowhere near complete, still a lot of work to do, then we might try to steer them in the direction of one of the systems that we're working with anyway.

Andy Webb: With the guys on the JV side, you mentioned earlier that you're expecting them to be way past the idea stage - you're not expecting them to have a real-time, real-money track record for the model, or are you?

Asaf Arditi

Asaf Arditi

Ron Hertshten: No. Obviously it's a preference if they do, but we realise that by that point they'll have less of an interest in working with us. It's not an absolute requirement. Some of them do have something. We have someone who had a model that worked really well in his local market, so there was some track record, but he really wanted to grow it on a global basis, scale it up. That was a track record, yes, but it was very limited and it wasn't on the markets he was looking to trade.

Andy Webb: With the people who come to you: do they tend to fall into categories in the sense they're either looking for access to capital, and/or cost-effective access to markets, and/or they're looking for the extra tweaking and additional IP that you can add? Do they fall into distinct categories, or are they a mix of the three?

Ron Hertshten: So far, they've tended to be a mix of the three, but with more emphasis on the first two.

Andy Webb: When you're looking at running one of these live, how do you work out the amount of capital you're going to allocate? Is it very model-dependent, or do you have a generic starting point - we'll put x behind it for so many months, and if it's going well after that, we might increase the amount to 2x?

Ron Hertshten: Exactly that. We're looking to start relatively small, to make sure that the model is working the way we're expecting to work, to make sure it's getting the results we're looking for. If it is - and this is why we have a strong emphasis on only models that are scaleable - we increase the size of what it's trading.

Andy Webb: On risk: are you finding that the outsiders coming to you for JVs have already got the risk aspects of their model pretty well nailed down, or is that an area where you're having to make significant extra input?

Ron Hertshten: It varies. We have a strong risk team, a lot of guys working on risk. Our requirements are often over and above what the algo would naturally have. So we make sure that the risk is appropriate for us. We put a very strong emphasis on that.

Andy Webb: With your own two models, how has the firm decided to allocate capital for those? Is it, like, your new automated area and you have x million to play with?

Ron Hertshten: We started off pretty small. It depends on performance. When it started - the one that trades in the STIR market, to use that one as an example: it started trading a few lots here, a few lots there, it was very small. We have hundreds of traders and we're recruiting more every year. This is something very regular for us - opening a new account, managing someone growing slowly.

We're very used to it and we didn't make a huge distinction in the amount of capital needed. It manages its position, it trades much more volume, much more frequently, but the idea behind it is not that different from a human trader who trades the STIR market. In terms of capital allocated, we didn't see a need to do it a whole lot differently.

Andy Webb: With the STIRS, presumably, because you're trading spreads the whole time, you haven't got much outright risk; therefore, with your margin offsets between the calendar months, you're not necessarily…

Ron Hertshten: It's not just spreads. It almost always has an exposed outright position.

David Lerner and Raphael Swerdlov

David Lerner and Raphael Swerdlov

Andy Webb: Ah, okay. With recruitment, you say you're recruiting new traders every year and it's a regular cycle. In view of your new automated business, rather than just recruiting people you're going to train up as traders, do you think you might be - maybe next year - taking on people who are quants, first and foremost, who you might dedicate to this business rather than them becoming generic traders first, or do you feel that the trading experience, going through that learning process, is integral to anything they might be able to offer you in the automated space?

Ron Hertshten: That's a question we've debated a lot and we still haven't come up with an answer. We're probably going to try both approaches, but we don't have a clear answer to which way would work better. There is a lot of value in exposing someone to the markets. They understand what the markets really look like beyond the numbers. But on the other hand, is that really necessary? Could someone get to the same results faster by not going through the process? Also, maybe someone who's very successful as a quant wouldn't be a good trader. We possibly lose someone who would be a great quant guy.

Andy Webb: In a sense, the trading experience, for a quant, is the reality check. You can look at data, crunch numbers, optimise models, do everything right, very rigorously, but you can still miss something. You still have the potential to do something, and a seasoned trader would still just say: you don't do that in that market on that day - and be right.

Ron Hertshten: I agree with you, but if you have the development done in teams, you can have the quant guy and the trader.

Andy Webb: That was exactly where I was going. I was wondering whether that would be a model you'd consider. You can then get the pure quant expertise, the model nailed down tight, really, really fast, and then the reality-check guy who says, come on…

Ron Hertshten: Without ever trading the markets, I think it would be really difficult. Because there's stuff that, if you've traded for a bit you understand it; but if you haven't, you might not. For a quant, you might see the market disappear from your screen, and you might not know that just before non-farm payrolls, everything goes. You might give it some other meaning. The trader would be able to explain those things away quickly. The model that we're envisioning is more teams.

Andy Webb: More teams. But you'll give the quants some experience without trying to turn them into full-blown traders?

Ron Hertshten: Because we're a trading group, we want everybody to have an understanding of that - we even have our HR people and our finance people trade, even if it's only simulated for a short while. One thing about that, if a trader goes running to an operations guy with a problem, the ops guy doesn't just promise to look into it after he's finished his coffee - he understands how nerve-racking it is.

Andy Webb: That's true…

Israel Arviv

Israel Arviv

Ron Hertshten: I think that could probably apply to quant guys as well.

Andy Webb: Any projections on how your automated business is going to grow?

Ron Hertshten: We don't have any projections. That said, I think it will grow into being something very, very big. It will be a major part of what we're doing. There's room for the human side alongside the algorithmic trading. We're in a fantastic place to be, where we can have the two developing simultaneously. There's a lot of spillover, and we've seen it already happening. Part of our challenge is to manage these separately but link them together as "hand-in-glove" complements to leverage our manual trading. The algos might notice something in the market, they all of a sudden get a feeling that something's different, and they yell it out. The human traders like that. Sometimes, it works the other way round. The point-and-click guys tell the algos that this, this and this is happening, and right away they make some adjustments in their parameters.

Andy Webb: When the automated models are running, who is responsible for managing them and making sure that nothing goes horribly awry?

Ron Hertshten: The team that runs the algo; they're sitting in front of it. Then we have myself, plus risk guys in a few locations all looking at the same stuff at the same time. Then we have some automated systems as well. So we have quite a few eyes and some systems watching at any given second of the day. It's not one person. In addition, our clearing arm adds another and independent layer of comprehensive risk management controls.

Andy Webb: Does that apply with the internal and the external alpha models? Is the same team watching it in actual execution every day, or are there people in different offices watching different models?

Ron Hertshten: People in different offices watching different models.

Andy Webb: So this really does sound like a potentially very globally distributed business; as much as your manual trading, perhaps.

Ron Hertshten: Probably more so. When we think about where we might want to have a point-and-click proprietary office, there has to be enough population, with high enough education, and so on. If we're talking about three guys who can write an excellent algorithmic model, they can just as easily be in the Seychelles as anywhere else. Chances are most of them will be in the locations that we're physically in, because that's where we have the connections, that's where we'll get to know them. But this could be very much more spread out globally.

Andy Webb: Do you think that you will be, in a year's time, recruiting people with more the characteristics of a quant for this automated trading group? Granted that, as you said earlier, you'll probably put them through some kind of trading experience.

Ron Hertshten: Yes, I think so. It's a special skill set that they need. We probably will be looking to bring in specific people for roles that we have.

Andy Webb: With the external JV partners, do you have a recruiting drive to find them, or do they just pop up anyway?

Ron Hertshten: At the moment, the ones we've been working with so far, we've just come across them. They might have approached us, or we've approached them, because we knew them through prior relationships, for example. That said, we're looking to change that going forward. We absolutely are going to be looking more actively to find these guys, and we're going to have to think more creatively about how we get access to them. We've developed a broad set of networks in the countries in which we're based, and we're aiming to systematically use these networks to source and vet opportunities.

Ron Hertshten and Jonathan Horowitz

Ron Hertshten and Jonathan Horowitz

Andy Webb: Do you publish a basic list of criteria for JVs?

Ron Hertshten: Maybe it's worth preparing such a document. It's been done on a very small scale so far, and I'd say that we're all in favour of hearing new ideas, even if we don't go ahead with all of them. We don't want to define a set of criteria or hang our shingle out online. We prefer to work through our networks to expand our exposure to interesting opportunities in a "no preconceptions, eyes open" fashion. I think this is a more open-minded but high-value approach.

Andy Webb: With your in-house and external models, how do you handle the issue of model decay? Have you had the issue of junking, or perhaps recalibrating, models?

Ron Hertshten: Most of them have been running for quite a short time. We haven't really faced that. The one that's been running the longest, a year and a half, has been improving all the time. They're constantly working on it, it's not a finished model, but no, we haven't faced that issue to date.

Andy Webb: Even though it's been live with real money for eighteen months, it's still a work in progress? They're still adjusting it although it's live?

Ron Hertshten: Once it's live, that doesn't mean they stop working on it. They're still developing it. If we waited for it to be optimal - 2011 has been a terrific year and we would have missed out on that. So we run it probably not at its optimal size, but we find it very much worth running. They guys who built it do believe that at some point they'll finish.

Andy Webb: They always say that! Developers…

Ron Hertshten: Yeah, I'm a little cautious…

Andy Webb: When you initially build the models, are you trying to make them as self-adaptive, self-optimising as possible, some of the logic based on volatility, liquidity, whatever, or do you accept a certain number of fixed parameters? Are you constantly adjusting just to change a few fixed numbers?

Ron Hertshten: Again, it depends from one model to another. Ideally, we would like a model to be self-optimising, but it doesn't have to be.

Andy Webb: With the two models you're running in-house: when you developed them, were you intending them to be complementary, or were they two totally discrete models that just happen to work for you? Do you have an expectation that when one is underperforming, the other will compensate?

Ron Hertshten: Interesting you should say that. We've noticed that each model is very, very good in a certain type of market. We look at it more as discrete models. From a group perspective, I am looking to see that we're not too exposed to one type of model. I'm looking to diversify across a bunch of strategies, but each model is a discrete situation.

Amit Ashkenazi and Amir Pertz

Amit Ashkenazi and Amir Pertz

Andy Webb: As your business grows, and you take on more JVs from outside, you might get into a situation where that gets too difficult to control? If 2012 is very good for trend-following strategies, say, at the beginning of 2013, you might find that you've got a lot of potential JVs offering trend-following strategies.

Ron Hertshten: Right. We're looking at the performance of each model as a discrete case, but I am looking at performance overall and I don't want to have too much exposure to too many similar models. You can have a momentum-following one that's based on weeks and months, and you can have one that's based on minutes and hours. So you have to look at them more carefully than just giving them a generic term. It doesn't matter what we call them; the question is, how are we expecting them to perform in a given set of market conditions?

Andy Webb: From your personal perspective, at an oversight level, you're looking to benefit from diversification by model type, but also from diversification by the individual application of the model, which could be the time frame.

Ron Hertshten: Yes, exactly.

Andy Webb: When you're looking at that from an overall perspective, what are your metrics? Sharpe ratio? Sortino? Or proprietary stuff?

Ron Hertshten: Mainly Sharpe ratios. We're also looking to develop some of our own in-house measures of performance. It won't ever be just one number; it will probably be a combination of different metrics.

Andy Webb: Do you have an over-arching performance requirement, or desire? What are your key points; do you rank smoothness of summed equity curve above outright P&L, for example? What are you looking for?

Ron Hertshten: We're looking for more of a stable model without too much volatility in its performance. However, if it's one model - if we build a portfolio of models, volatility in one model becomes less important. That's why we're looking to build a few in-house models and mainly focus on the joint ventures to come up with great models that we implement. If one model behaves with a lot of volatility, that might be great for us in the long run, and the other models might compensate in the short run.

Andy Webb: This has been such a fascinating conversation. Ron, thank you so much.

Koby Batito

Koby Batito