I am Santiago Braje. My background is in banking and finance from when I started my career. My family was involved in finance, I grew up within a family where my Father owned a car loan company in Argentina, he was the second generation running the firm. So I was exposed to credit from a young age, my first actual paying job was to work with my Father in the office.
I then studied Economics at University, went into banking in Argentina with Societe Generale initially and then quickly moved to Citi. I relocated to London to study for a Masters degree in Operational Research after which I relocated back to Argentina and worked with Citi again. I had a few more years with Citi and then had my first short experience as an entrepreneur, I founded a consulting business with some colleagues from Citi to work on debt restructuring projects. As there were so many crises in Argentina, this was the one around 2002. Virtually all companies, apart from commodities exporters, were in trouble with their debts so all banks were heavily involved in debt restructuring. I worked on both the bank and debtor side of the business.
After that I came to London to work for ING in the Credit Trading Business. I started in what was called “Exotic and Illiquid” at the time. Over time I had several different roles within Credit and Fixed Income, over the full spectrum from of liquidity to the illiquid extremes of structured derivatives, leveraged notes and those sorts of products. As I moved up the ranks I took on more responsibility, in 2009 I started looking after the flow business. By the time I left last year  I was running the Credit Trading business globally.
Last year I left to found Katana Labs. Katana Labs is actually a company that started as a project within ING
Katana Labs combines what have been my two main professional interests: finance, in particular fixed income and modelling and decision-making tools. How do you make decisions? What information do you use? How do you put a framework around it? How do you model the world within your decision space? How does this help you make better decisions? Not just in terms of automating things but in providing better insights.
Originally, I studied Economics and what drove me into Operational Research was that interest in the modelling side of things. Not modelling generally but rather modelling practical problems and particularly decision-making problems. That’s where my interest arose in what Katana does.
The project itself started as an internal idea at ING. We had an initial project to help the traders in my team with their day-to-day decision making around liquidity provision, how to quote on electronic and manual RFQs, how to do this faster and more effectively. We developed a tool using Machine Learning to help with that task.
Then, out of that we had a spin-out project. We thought we can do this, we have developed a good approach to how to bring Machine Learning into an actual tool that you can use. We then decided to use this to help our clients. We started first with PGGM in the Netherlands to identify the problem that was their biggest pain point and where we can make the most impact. The intersection of what was relevant for PGGM and where our approach, expertise and knowledge could have the most impact.
The first prototype was very much a collaboration, we did all the work and they provided their feedback. So we iterated through several prototypes to a working product over about one year. Then we decided the product was mature enough to scale. We also decided that we would be better placed to scale the business outside rather than inside of the organisation [ING]. We decided to spin-out and Katana Labs was formed in November 2019, so we are just over one year old.
It has been interesting to launch a new company into this world, in this year that we have been having. But this is also affecting established players, the playbook for what you would do in a pre-Covid environment, you have to completely throw away. You have to think again from scratch. How did you get awareness in a pre-Covid environment? You had to go to the trade shows, in the big events, have your stands, give your presentations. You had face-to-face meetings, built connections and built relationships, that was very much how the sales process worked.
For us, we launched the product in January. For the first two and a half months we were working in the pre-Covid way. We were very successful, face-to-face meetings and getting to a point of sale on a clear path. The sales cycle was not very long since the product is very light in terms of adoption, there is no integration with systems or data, it’s web based so it’s very easy to start using it.
One of the biggest issues we have found as a new company has been the change, from sales meaning flying around to meet people to more of a research project to find the right people with whom one must speak.
With Covid everything changed. It may change the economics of the sales process in a positive way, but I think that has not happened yet. The first effect has been that the traditional sales channels are no longer there anymore. We have noticed that there has been a cycle, right at the beginning of the pandemic when people started working from home you had a sense of inertia, the conversations moved online, Zoom etc. Then people started to get busy with establishing a remote working pattern, spending a lot of time on Zoom calls but becoming fatigued with that. With regular calls with your boss, team, suppliers and so on people didn’t have so much time to start new relationships. I think we are coming out of this slump now as people are settled in a new way of working and are back to doing business in a normal way, albeit with constraints and restrictions. So we may end up seeing this as an advantage, as you will be able to do business without flying around.
The long-term impact remains to be seen.
Over the last couple of months people are getting on with things, especially with the end of year looming and planning for 2021 taking shape.
What does your firm do?
We do pre-trade analytics, we focus on relative value. We deliver relative value insights and analytics for bonds. What is new and distinct about our proposition is that we tackle this problem in a systematic way. We have built from the ground-up a system that is dedicated to solving the problem of monitoring and identifying relative value opportunities in the bond market. When I say we do that systematically we use machine learning to analyse the market. We look at every pair of bonds that can trade within a given universe. Analyse each pair with a machine learning algorithm and then identify where there is a relative value opportunity and changes in relative value that require attention.
The underlying thesis is that if you are involved in the bond market, whether as a portfolio manager, trader, analyst, salesperson or advisor you care about relative value, whether this is explicit or implicit. VA value is always relative, nothing is cheap or expensive on it’s own, it’s always with reference to something else.
As an example, for houses, you don’t make a judgement on whether a house is expensive or cheap without looking at the neighbourhood, square metres and other characteristics of the house, the price of the house in the past and the price of other houses in the area over time. You have a frame of reference to make a decision on what is fair value for a house.
In the same way we do that for bonds, that’s one of the key things we believe, value is always relative. Another thing we believe is that opportunities are anchored in some change, when the world changes in some way and prices don’t or if prices change and that’s not justified by some change in the world. In other words, the relationship between prices and fundamentals changes. Given that it makes perfect sense that you want to observe and understand what has changed in the world, os if your world is investing in is US Credit you want to know what has changed in US Credit. But, you don’t just want to know what has changed. You want to know what has changed in relation to what. And effectively that’s what we mean when we say that relative value has changed. This is something everyone does, if you are an investment professional you are very, very likely to do this. I haven’t found anyone yet that does not have some way of making that judgement of relative value. Either a very simplistic way of looking at an index versus a particular bond, or you may look at a subset of bonds such as an industry. You may have one of more of these methods. In all cases there’s an assumption – some way of looking at a reference to a price.
We take a scientific approach, we make no assumptions about what is relevant to compare. We look at the whole universe, analyse each pair without any constraints of what should be looked at and what should not be looked at. And that’s where the Machine Learning piece comes in, as a machine is well suited to this heavyweight analytical processing, whereas a human will have a limited capacity to conduct this analysis. At the moment we cover 30,000 bonds throughout the bond universe, US IG Credit, EU IG Credit, Euro Govies, EM Global, SSAs. This means when we look at pairs we have hundreds of millions. With this amount of data we can train the algorithm to look at two things:
(1) We identify what are relevant factors that should be analysed – we look for bonds that have cointegration over time, but also are stationary. One problem with looking at large sets of time series data is that you will see spurious correlation, correlation will appear when there is no real link. One way to solve this is to impose more strict restrictions, you don’t just look at correlation but also look at stationarity. This reduces false positives significantly.
We started with a Bayesian perspective of complete rationality and then bring a judgement to that data. We found this was too scientific, that there is valuable insight in considering how people look at the market – in similarities in industry and country. So that when I see a dislocation in price of one bond versus another and it appears within the same industry category this makes sense to me as an opportunity.
Hence we added:
(2) The static characteristics of the bonds, industry, country, rating, maturity and so on. We then built a similarity score using machine learning on these static characteristics.
This means we have a system that can measure dynamic similarity and static similarity. This can then throw away say 99% of what we look at. Within the remaining 1% we want to identify where there are dislocations that are relevant.
We provide an application that gives those insights, from say 10 million pairs you look at, you may see a couple of hundred dislocations on a given day.
That’s where we are now on the product. What we are working on now is to make the workflow as natural and efficient as can be. We are working to find an interface that can leverage the power of machine learning that can be consumed and interpreted by a human brain. A lot of this has to do with the flexibility with how you interact with the data, how you visualise the data, how you bring in your own data and your own portfolio. We started with a very agnostic perspective, we learnt that people wanted to know what is happening with the portfolio I they already hold, what is happening with the watch list of bonds in which I they have an interest. They may even have preconceived ideas about pairs of bonds that they are interested in or bonds they want to compare. They want the tool to give them those insights also. So we need to merge all of these features into more of a hybrid – between a purely quantitative systematic trading strategy, where you write the code and just let it run, and a purely human based non-systematic strategy. We want to find the point where these things come together to give you the best result. It’s a really interesting challenge, not an easy thing to solve. But we are happy so far and so are our clients, so we are heading in the right direction.
There is so much more that we can do, at the moment we are really just scratching the surface of what we can do.
What made you pivot from a successful career as an Investment Banker and joined a newly incubated fintech start-up?
This was an interest of mine for a very long time, so the idea of starting something of my own, of creating something from nothing was a long-term interest.
The second thing is that I found this is where I was most interested; an in the intellectual challenge but in a very applied way. The idea of creating the models, the frameworks that help you create better decisions, use using some clever maths and statistics and ultimately reduce the problem to something where you can make better decisions.
The third thing was seeing what seemed to be secular trends in the market – one is the continued growth of the bond market – increasing issuance of bonds and a market becoming more complex generally. On the other hand the incredible pace at which machine learning as a discipline has been evolving and the capabilities of machine intelligence generally.
It seems pretty clear to me that the future will have more and more machine intelligence in the fixed income world. So, in terms of career, the future lies more in what I am doing now rather than what I was doing just over a year ago.
As a Fixed-Income Fund Manager who is not tech-savvy, what can you do for me?
If you have Katana you can upload your portfolio and watch-list. You can then see if any of those bonds are cheap or expensive against whatever else. You get a way of navigating the whole market space from the perspective of what you own or watch and can then see where you could extract value. It’s saving you time, you can go straight to the things that deserve attention. You will see where there are changes that you may have missed, things you may not be watching and you will see this when it happens.
Blockchain / Distributed ledger technology – in use with your firm? Hype or real?
We don’t use DLT. There is something very powerful at the core of this technology. The use cases are not yet string strong enough to be ‘game-changers’. Do we think there is a problem of trust within Fixed-Income? I don’t think so. Not trusting a central institution is not a real concern. There is a core proposition of a replacement of trust in a central institution with decentralisation. I think we will see a future iteration that will have a big impact, but not so far.
Cloud - in use with your firm? Hype or real?
We are cloud native. I mentioned two secular trends earlier, a third is the rise of cloud computing. That’s not only the ability to spin up machines and scale computing power quickly but the whole ecosystem of architecture, open source, development methodologies and the speed at which you can develop and deploy.
We built our first prototype in six weeks and it was being used by a client then. At the end of the development you have a cloud platform that is robust and well architected, you can do releases on a weekly basis. The ability to develop and deploy on a weekly basis is a true game-changer. The software we use at home works in this way, financial technology should be the same, frequent enhancements and releases.
The new generation of fintechs are all cloud native, adopt this or you will be left behind, on-premise is part of the past.
Fintech – is that how you describe your firm?
Where do you see technology impacting in the next five years on:
(b) Your firm
(c) Wider market
We have gone through an evolution in Fixed Income. An initial phase of information, go back to the 1980s where Bloomberg was the innovation. Over the last twenty years we have seen an evolution in connectivity, think of Tradeweb and MarketAxess. We are now entering a third wave of innovation around machine intelligence. The data is there, the connectivity is there. The amount of data and the velocity at which is moves s beyond what human brains can manage, so the obvious evolution is machine intelligence to create actionable insights. We mediate the data and connectivity with machine intelligence.
To look at this from the perspective of Katana, you should not be looking at screens and seeing numbers change, it makes no sense. You have four, six, eight screens full of numbers changing. The human brain is not evolved to do this. Let the machine do that work, they do that work really well. Then you bring your insights, expertise to this to make decisions.
To me, this is obvious. The only thing that is holding people back is that people are used to doing things in the way that they do.
The core value we provide is that we can drive the attention of a portfolio manager to what needs attention. Otherwise you just spend your time being driven by external inputs that may not deserve attention.
Post a Comment