What does your role as head of strategies and data science involve at BBVA?
As a data scientist, my role naturally involves the (rare) set of hybrid skills required in this domain: the business mind-set of an economist, the accuracy and creativity of a scientist, the insights and risk expertise of a trader and the independence and techie acumen of a developer. All these skills are then fused to re-structure the current e-commerce and e-trading strategies, fine-tuned and shifted to a new level (a learning-adaptive one) and exploited globally across desks and assets (synergies exploitation).
On the other hand, as well as being responsible for the whole change, the role revolves about education, incentives-alignment and motivation. Without education, possible partners across your colleagues can become antagonists, as being humans we tend to fear the unknown. The incentives-alignment is crucial as it avoids delays at the price of some political give ups (not my main worry). Without motivation there is not the (extraordinary) energy required when adapting to changes in environments.
Could you tell us about how your long term development (academically and in the commercial sector) have led to this position? Do you feel that each stage has been an important part of the process?
Good question. My long term development has definitely been crucial. Along my career I have been swinging between the two beasts; academia and industry. The more fascinated I was about the former the more excited I became about its application along the latter.
Since I have done a lot of things in a short period of time, it may be too long for me to mention the role of each stage. However, I will go through the highlights to show how all the stages are linked
1 While at college I started up an econometric department of gas demand prediction at a large company. I did well and was offered a position to head a new team of applied science; a role that I used to call “inSCIder”. I was fairly tempted by the idea of mixing academia and industry on a daily basis; however I felt I still lacked quantitative skills to take it up to the next level.
2 I then accepted an internship at a top MSc in Economics and Finance linked to the Bank of Spain. This turned out to be somehow too theoretical for what I had in mind. I was lucky as my dissertation was quite a success; it revolved about the application of quantitative tools upon fixed income data. I understood back then that finance was a great field to gather data and apply science. My dissertation grabbed the attention of a top academic in Asia and I was offered a fellowship.
3 After the MSc I had the chance to apply to several places, but I decided to accept an offer from a risk consultancy as I thought that every trader should begin with mastering the risk management domain.
4 After a year I moved to Japan where I learned new ways of approaching finance and a working discipline that I do admire. Whilst in Japan I was contacted by Santander (where I had previously worked as a risk consultant) to start up yet another department in my career. A department sitting this time within a Trading Floor (fixed income). The quantitative skills were a must so I thought it was the right chance for me to apply what I had learned so far. It turned out that the business pace didn't leave much room for applied science (just a bit of coding here and there for efficiency purposes). I instead learned how to work under huge pressure, how to reach agreements with troublesome departments and more importantly, how to be aggressive whenever it was optimal for the project.
5 I managed to stand out in this new area and despite being in Spain and in a commercial bank I was able to grab the attention of JP Morgan and Morgan Stanley. The former offered me a position similar to what I was doing and the latter a position in a different asset, with more pressure. The latter was also only a year-old-project which implied higher risk and again, more start-up mind-set. Moreover, it also involved the execution of trades through algorithms yet another field that was new to me. I’m sure you can guess which one I picked.
6 This is how I found myself working with algorithms again. I came up with a few tricks here and there which helped us quadruple the P&L in less than a year by being more efficient than our competitors. I wanted to progress forward as using the algorithms smartly was not enough; I wanted to design them from inside. This is how I then took Lehman's collapse as an excuse to leave the industry and go back to academia to learn machine learning, a handy discipline that I lacked.
7 I was fortunate as back then the UK PhD Centre in Financial Computing was being launched and I was offered yet another internship. The director of the centre, Philip Treleaven became my first supervisor and Bob Jenkins my second supervisor. Finally, I managed to grab the attention of the CEO in Europe of Renaissance Technologies (the financial agent I admire the most) and somehow had the feeling of having reached the top level in finance that I always dreamed of (no, I don't measure success by money or seniority). Anyhow, with this team backing up my research I was soon offered several positions at several tier-one-investment-banks and top notch hedge funds.
8 And surprisingly, I turned them all down to join BBVA, another commercial bank (as was Santander) as I thought it gave me the best framework to innovatively start up yet another department. In a nutshell, I have been stubborn about the need to have expertise in applied science within the industry and about he need to become an “inSCIder”. I have taken risks and paid a price for leaving the industry every so often. Lucky me, applied science is finally taking off and over.
How important would you say data science is at BBVA?
As I mentioned above, before joining BBVA I had the chance to join some well-known investment banks and hedge funds which I turned down in order to get a placement here. Some thought I did wrong; they were only valuing the money. Others (related to my dissertation... food for thought) agreed that it was the smart move; they were valuing the career opportunity. I say opportunity because the placement that I took back then had a much tighter scope than the position I currently hold. However, it had the right upside: the chance to gradually prove the value of data science in the future of trading. The opportunity was in fact two fold. It was located in the right unit - at the forefront of the trading innovations and within the most competitive asset. More importantly (and here is where I wanted to come), it was in a bank with the right philosophy to foster the change. BBVA is led by a President who is one of the pioneers in commercial banking innovation. At the same time he happens to have developed his career within the investment banking side of the bank. Being in this position means that you will eventually be able to start doing big things.
Hence, to your question, the role of data science at BBVA is large enough to inspire the whole chain of value. Large enough to attract rare individuals who can set up big things along the bank.
Could you give us some success stories about your projects? How was success measured?
It is true that when you try to innovate, it is not always easy to measure success. Sometimes, more often than not, success is being able to provide the right answer to a troublesome question on the fly, in front of the senior management (and using the hybrid skills mentioned above). Sometimes it is the other way round; being able to provide the right question. Success can also be achieved if what you forecast that will happen if what you propose to be done is not deployed, happens. For that reason, successful stories are more than mere numbers and much more difficult to measure.
Working in finance though means that I am lucky enough to have stories that in the end boil down to numbers, most of which I cannot of course tell. An eloquent one though, which I love as it was one of the firsts, was the need to provide a very sensitive execution for a specific client. Luckily enough I was quite advanced back then in the process of education mentioned above. At that stage there was not an official role for me across teams but a latent one. I personally decided to take full responsibility for a task which belonged to a different team; I believe that, going back to the previous paragraph, this counts as a success. But in this case, I can dig deeper and add that after taking a savvy shortcut in order to tactically approach needs that we strategically needed but didn't have in place yet, I managed to reach a 0 basis points deviation. That's an A+ and it served me to keep motivating the need for a change.
How difficult do you find it getting top level buy in from the business leaders? How frequently do you feel your ideas are adopted and made central to business decisions?
Interestingly enough, real innovation is not glamorous. It is risky, rough, and frustrating amongst other things. Not in vain I have been freezing my PhD thesis for a couple of years now in order to have an academic hedge in case things go wrong, even though I already published it as a book back in 2012. Innovation is all that because even though in reality it is not that difficult to get top level management, buying rock-the-boat ideas (of course, when you already have some industry credibility) it is however extremely difficult to have a smooth access to it through the support of your immediate superiors and colleagues. As I mentioned before, it is the fear to the unknown that drives a defensive behaviour from the latter and squeezes any chance for you to speak to the senior management. The only solution I have come up with to overcome this situation is education. My advice is to find the smartest mid-managers and re-iterate the education upon them. Not only will they give you access to the senior management, but they may also pave the road for you. Once you have visibility and everyone understands inside-out what you are proposing, you are almost there.
With regards to how frequently I feel that my ideas are made central to business decisions, I have to confess that for the last couple of years I cannot complain whatsoever. In fact, they seem to be made too central now for me to avoid PowerPoints and endless meetings.
How would you describe the data science market as a whole – does your work sit right in the middle of the market?
I look for more from a Data Scientist than the average person I would say. I have noticed that the industry is assuming that in the end it boils down to database management, a quantitative approach to its exploitation and a large dose of common sense. Whilst those are necessary conditions, I would not class them as sufficient conditions. In my view, and it is something that I subtly implied in the previous questions, you do strongly need expertise in the field. A data scientist is the right mixture between industry and academia as the best ideas lie behind the expertise of the professionals. It is in the business acumen where the more robust patterns can be found what in turn shall be core to the (always difficult) weigh off between the theory and data driven worlds. However, they sometimes have the information but cannot relate it to the solution of the quantitative challenge because they lack scientific sensitivity.
As such, I still see the data science market as an interaction between two parties: professionals and scientists. For this reason it can be perceived as sitting in the middle of the market as you mention above. However, I would expect it to be an upper layer sooner than later. Data scientists should be able to see the whole picture from above and be fairly independent.
Are there any Data Scientists/ Data Science projects (could be start-ups or other departments, etc) that you really admire for their work in the field?
This world is full of smart guys doing really good things and I admire many different people. If I had to give an example, I would say both Renaissance Technologies and SciTheWorld.
It is likely that Renaissance Technologies were the first team to use big data in the world. The team consists of very smart academics lead by James Simons, a Phelps medal (also known as the Nobel for maths), who have gained a robust edge in the financial markets for tens of years by smartly exploiting patterns that others were not looking at (I expect the usage of intraday data to play a large role on that).
SciTheworld is a project that I am starting up with a group of idealistic people who believe in data science, low-cost innovation and open knowledge. We are building up a database of insiders with a scientific flavour who are keen to define the projects along with academics (outsiders). More or less, we'd like to industrialise what I have learnt from my experience here in order to pave the way for others with a less hybrid profile. We want to show the world the big things that can be triggered through the right matches. Unfortunately, we do not have much spare time to put all the strength we wished to push into it; we will see what finally happens.
What trends/changes have you noticed in the industry?
I have noticed that there is a worrying amount of non-scientific people forcing their role in the area. Being science an entry barrier it is easy to make a pitch full of technical words and wow clients or colleagues. If they do not understand what you are talking about, you don't need to understand it either. More eloquently, I have seen people calling algorithms to thumb rules.
This is bad news for the whole industry because at some point the crowd will start claiming that data science is a bluff, when the reality is that the people being selected are the bluff. I encourage every data scientist to try and pick up on this.
What advice would you give to somebody trying to break into the data science industry? What do you think makes a good data scientist?
Do not rush into it. The role of data scientist is here to stay and it takes time to gain and develop the right set of skills. Therefore, take your time and acquire all the skills you can get. That will make you stand out in the long run. In a nutshell, a rich background is what makes a good data scientist.
What predictions do you have for the Data Science market in the next five years?
For the next five years there won't be many profiles that gather both the business acumen and the scientific expertise. So we will have to rely on the interaction between the most quantitative professionals and academia. Hence the need for free matchers such as SciTheWorld.
Do you have any favoured technology stacks?
I am a big fan of Python and R-CRAN. I believe in the future of Infobright, however I am still reluctant to use it in production. And I was not too pleasantly surprised by the fact that the latest version of RapidMiner does not seem to be free. I therefore definitely prefer Weka.
It is important to note that depending on the stage in which you and your project are within your company, you may have more or less budget restrictions to start delivering and scaling the project. In these situations free software then becomes a must.
The next question to then arise is whether or not I like non-open source tech; the answer to which is that yes, I do. This is mainly because within institutions, it sometimes becomes a must at a production level. That is why I tend to favour open source products with an enterprise edition.
Could you tell us a little more about your book? Which areas do you feel could do with further publication within the Data Science field?
"Trading 2.0: Learning-Adaptive Machines "is simply the book I wished I had when I decided to move onto this side of the industry. It is all about the insights upon the latest trading paradigm across the main domains in finance: execution, market making and proprietary trading. Ultimately it revolves about what I care the most: the out-of-sample robustness of the models. For that, I simply propose to launch a bunch of robots across the calibration universe to then analyse what they learn about trading and compare with what the trader knows: reaching this way a weigh-off between the theory-driven and data-driven worlds.
There are currently no limits to financial innovation, and I am sure that there are many ways to democratize a good part of what is being developed in advanced finance. For example, I participated in the adaptation of a very well-known trading index, the Sharpe Ratio, so that it could be used by entrepreneurs. The merge was called Sharpyme (could be translated into Sharpreneur) which helps entrepreneurs gain real transactional data insights in order to understand the risk embedded in the different business opportunities that they are considering.
This is just an example and I am sure that there are many ways of helping the industry take the right choices in terms of risk if we are creative enough to transfer our knowledge to the 'real economy'.
Are there any methods/ machine learning processes that you have favoured and come to rely on?
Let’s say that I use an open-end range that involves from support vector machines and reinforcement learning to particle swarm optimization. And probably I use these not the way most of my peers do. Beware that as pointed out above, to me the most important bit is to find a merge between data-driven and theory-driven techniques; I would expect the largest value of the data scientists to be in that very fusion.
What would your response be to those that criticise Big Data/Data Science market?
Put your money where your mouth is.