In the world of fixed income trading, the use of automation is becoming increasingly prevalent. From algorithmic trading to machine learning-based systems, technology is changing the way that deals are struck and trades are executed.
Alexei Tchernitser, Director, Analytic Solutions, Quantifi and Ersel Korusoy, Executive Director, Standard Chartered Bank discuss the ways in which automation is redefining the front-office and transforming the fixed income market as we know it. They explore the benefits and drawbacks of automated trading, and consider what the future may hold for human traders in this digital age.
Ersel has worked in finance for 20 years. Roles held include Co-head of Fixed Income Electronic prop trading at Winton Capital Management in London, Head of xVA, Funding & Rates quants at RBS in Singapore and, more recently, Automated Fixed Income Electronic Trading at both NatWest Markets and Standard Chartered.
“If you want to make money today with quantitative strategies, you need to be an economist, a trader and a mathematician.”Ersel Korusoy, Executive Director,
Standard Chartered Bank
Ersel: In today’s landscape, automation has largely brought bid offer spreads in. If you show mid today, for example Tradeweb mid on German Bunds, your hit rate is something like 20%, which is barely enough for you to have a franchise. When I left NatWest Markets to join Standard Chartered less than a year ago, roughly 90% of Gilts, Treasuries and Euro Govies were being handled by automated systems. So, that is kind of where things are at the moment. If you want to make money today with quantitative strategies, you need to be an economist, a trader and a mathematician. You need to ask yourself, “What have I seen today? What kind of news has come out today? Is it going to be a trend following day or a mean reversion day?” Only by combining all of these things is it possible to have strategies that are profitable long term.
Alexei: So, you mentioned about 90% of the volume in government bonds is traded electronically. What do you think that proportion is for less liquid assets, like corporate bonds, emerging markets?
Ersel: There is a great deal of automation in corporate bonds. One of the largest products that we trade at Standard Chartered is corporate bonds and it is very automated, but it depends on whether you have an API and a broker. If you have a broker that offers you an API, then fine, or you can automate it otherwise. So, it is really a question of the market rather than anything else.
Alexei: But it is still substantially lower than the government bonds?
Ersel: Well, yes. If you are trading something where it is mainly voice traded, for example, Mexican bonds, you can’t really trade those with an API. There is a very limited futures market, but you largely have to do that by voice.
Alexei: When a sell-side bank receives a request for quotes, what criteria can be used to direct that request to a voice dealing desk or an automated, algorithmic route?
Ersel: So, if it is a standard ticket, for example, and somebody sends you a bond that hasn’t traded for an entire month and they want to trade T+5 or a very unusual settlement, obviously, you would not want to handle that with a system. But, if it is standard size tickets, standard bonds etc. you would do that in an automated way. You wouldn’t bother the trader with that.
Alexei: Do banks typically take into account client tiering or considerations like that?
Ersel: The main concern with client tiering is if somebody hits the market at the same time that they trade with you. Clients are allowed to front run dealers, unfortunately. So, by looking at the market impact, you price that into the spread. It is much better to do it like that, than to try to do it manually, because you have got people like BestX who can come along and say, “why are you being nasty to this client?” and if you don’t have the data to prove the behaviour, it is very hard to justify the tiering.
Alexei: One more question related to that. Do you think that the algorithmic and systematic market making platforms are resilient enough to deal with the market volatility we’ve seen in 2020 and beyond? Or do systems like this requires human intervention from time to time?
I don’t think automation will ever replace people. I mean, it is like driverless cars that still need people to tell them where to go.Ersel Korusoy, Executive Director, Standard Chartered Bank
Ersel: Well, I don’t think automation will ever replace people. I mean, it is like driverless cars that still need people to tell them where to go. What we are doing is building fourth generation tools for people to do their jobs. There will always be situations where your model is not working and if somebody does not know where the limitations of their models are, that is a problem. I think it is important to understand that every model has a limitation, and to know where those limitations are, rather than to try and build the perfect model and think that there is no limitation.
Alexei: In 2020, we witnessed some of the highest volatility, and you would expect dealers to switch off electronic execution systems and switch to more traditional voice trading. However, possibly because everyone was working from home, you saw the opposite effect. The volume of electronic trading actually increased during that time, which kind of shows a much broader acceptance of this mode of trading now. It is not clear whether a lot of this volume was handled by automatic execution systems or by people sitting on the other side of the platforms, but there is a clear trend there. The electronic execution venues actually made it possible for the market to continue functioning and recover from that situation. Was this your experience as well?
Ersel: Yes, absolutely. I mean, it is really important to understand where the limitations of your model are and that is the critical thing.
Alexei: What are the typical machine learning or AI models that people use to set up a systematic market making or algo desks?
Ersel: Well, I am going to say something controversial. Machine learning and AI is a great PR tool and people love to invest in these sorts of things. I think that the returns are very limited, certainly when it comes to looking for statistical arbitrage opportunities, graph trading etc. I don’t think it works. You can use machine learning to identify client behaviour and patterns in client behaviour. Whenever there are interest rates going in a certain direction, or certain things happening in the market, different clients will behave in different ways – so, it is useful for that. It is also useful for identifying strange trading activity and most of the Big Four firms can sell you a package that will do surveillance using machine learning techniques. So, whilst I think that it is useful from that standpoint, actually making money by predicting which way financial markets are going to go – I’ve never seen anyone make a lot of money that way.
Alexei: Yes, I tend to agree with this view because I think that you get multiple people doing this, and the arbitrage, if it is there, goes away very quickly. But do you think you can use machine learning and AI tools to automate the job of a regular human market maker?
It is a trader’s job to listen to what is going on in the news, to listen to his colleagues, to talk to the salespeople and to talk to the client.Ersel Korusoy, Executive Director, Standard Chartered Bank
Ersel: Okay, so if I can replace a trader with a machine learning model, I should sack him because he is not doing his job properly. It is a trader’s job to listen to what is going on in the news, to listen to his colleagues, to talk to the salespeople and to talk to the client. These are all things that a machine can’t really do. So, if the decision making process of that trader can be encapsulated in just data and feeds, then that trader is not doing the job.
Alexei: One of the reasons I asked that is because what I have observed trading emerging market bonds. During the period of increased volatility in 2020, many firms turned down smaller sized requests because the trader could not physically cope with the volume of requests that the desk was receiving. I think this is one of the reasons why banks are investing in technologies to supplement the traditional sell-side desks.
Ersel: Exactly. It is a fourth generation tool to help people do their jobs.
Alexei: Okay, in light of that, can I ask you about the skills that people are expected to have now in these jobs?
Ersel: There are the three things that are important. One of them is that they need to be mathematically competent. Secondly, they need to be able to code. I don’t care what programming language, it doesn’t matter. The important thing is that they can code. And finally, they need to understand economics. I think if you don’t have those three things these days, it is very difficult to be competitive.
Alexei: From my side, I just wanted to add that I recently attended the RiskMinds conference in Barcelona and listened to a talk by John Hull. One of the things he said is that he requires his MBA students to take a five week Python course, which is a prerequisite to one of his machine learning courses. It is quite incredible that MBA students are now expected to learn programming because that is a required skill when they go and look for jobs.
Finally, what are some of the challenges that you see for the immediate future in terms of the profitability of doing business in the market?
Ersel: As I said, market bid offer is getting tighter. As more and more markets become electronic, we find that the barrier to entry is lower, the bid offer gets tighter and it becomes harder and harder to make money. A lot of bond businesses these days are much like McDonald’s. You go to McDonald’s and it is the same price everywhere, so it is very hard to make money in that situation. There are still some opportunities. For example, in Italian BTP markets, you find that the bid offer spread is very tight compared to the risk that you are taking but there is predictability in that market. So, if you are clever, you can position your book in the right way and make up for the fact you are not charging enough for risk. For more efficient markets like Germany, you show mid and you are at 20%. You have got companies in the US that are trading things like US Treasuries. Unless you have an enormous churn, it is almost impossible for you to make money trading those markets these days. So I think what is happening is we are seeing a consolidation, to some extent there is a sort of a technology arms race for the high frequency guys. It is just becoming harder and harder to make money and the barriers to entry are getting higher for people who want to be profitable.