Interview with Eran Raviv, Ph.D., Sr. Quantitative Investment Analyst
About Eran Raviv, Ph.D.
Eran Raviv is an Econometrician in the Netherlands where he serves as a Sr. Quantitative Investment Analyst for a large pension fund manager. In his present role, Dr. Raviv monitors, models, and interprets economic and financial data to help clients make strategic decisions about their portfolios. He is also a researcher whose papers are published in such peer-reviewed journals as Energy Economics and Economic Letters. He publishes some of his work—along with tutorials on various data-related topics—on his personal blog, EranRaviv.com
Dr. Raviv is passionate about econometrics, statistics, and data science: his blog posts and tutorials have been featured on EconoAcademics.org and the data news and tutorial site R-Bloggers. He is also a tutorial author for DataScience+, a teaching resource for data scientists working with the R programming language. Data topics of personal interest for Dr. Raviv include applied forecasting, dimension reduction, shrinkage techniques, and data mining. Dr. Raviv holds a Bachelor of Arts in Econometrics from the Ben-Gurion University of the Negev; a Master of Science degree in Applied Statistics from Tel Aviv University, and a Master of Science in Quantitative Finance and Ph.D. in Econometrics from the Erasmus School of Economics at Tinbergen Institute.”
[OnlineEducation.com] Recent movies, news articles, and other cultural touchpoints made “big data” and “analytics” nearly household terms. When someone decides to research analytics jobs and programs, however, phrases like “data science,” “data analytics,” and “business intelligence” begin to emerge. Will you describe what these three terms mean, including how they do—and do not—differ? What kinds of careers are associated with each, and do they all require the same degree of technical expertise?
[Dr. Raviv] ‘Data scientist’ and variants thereof are simply contemporary names for the modern statistician. Over past decades, we have experienced enormous technological advancements. Capitalizing on this increased computational power, [we have developed] many new techniques. So different that there was a need to create a distinction between ‘Statistics’ and ‘Data Science’. The former bears connotation to classical hypothesis testing and regression fitting while the latter carries the current buzz and glory of ‘big data’ and ‘machine learning.’ This newly created jargon has yet to stabilize, but I think most who practice it (however you want to call it) don’t feel any need to classify themselves one way or the other.
What everyone is after is to extract information and knowledge from data, whether it helps us to understand how something works or helps us in prediction and forecasting. For me, the distinction is simply artificial. The notion that if your data analysis leads to improved business operation, or gaining a competitive edge, then your field should not be called ‘data analytics,’ but rather ‘business intelligence,’ is preposterous. All those terms–along with some others not mentioned–are strongly intertwined. You see that in the practice, and you see that in academic research.
Kaggle is a website which organizes forecasting competitions. The winners of those competitions come from a large variety of disciplines. It is utterly wrong to think that someone who was trained in business intelligence will undoubtedly win a marketing competition. I often find myself reading machine learning journals as they are often referenced by classical statistical journals, or econometric journals for that matter. The only distinction I make is between analysis for the sake of inference and analysis for the sake of prediction, but it’s all still statistics.
If you are proficient with the techniques, you can integrate into almost any modern organization I can think of. From “old” industries like car manufacturing to “new” internet-based industries. Data availability is never the problem. Data storage capabilities have advanced much faster than the ability to analyze the data.
[OnlineEducation.com] Employers and colleges sometimes divide analytics into more specialized categories, like predictive, strategic, applied, and quantitative analytics. You are a quantitative strategist. What does this mean? How is quantitative analytics different from the other specialties?
[Dr. Raviv] I imagine a political advisor is someone who must be very analytical in his/her way of thinking, but there is nothing quantitative about it. Quantitative analytics means that you back your opinion up with data. Sometimes it is useful to combine the data with models, and this is where statistical and econometrical training helps. You need to know which tools and methods are available, and which are suitable to adopt.
Nowadays, it is fairly easy to apply a very wide range of models. It is a bad idea to do that without appreciating what is it that you are doing. You can count on the computer to spit out a number, but you must understand what it means. Related to that, there is so much data out there that sometimes simply singling out the important bits may provide sufficient evidence. The word ‘strategic’ has to do with the level in which I operate; it has nothing to do with some exotic super power.
[OnlineEducation.com] Reports suggest data expertise is in high demand across many different industries—not just politics, sports, and technology. You hold a Ph.D. in Econometrics. What is econometrics and what type of person might choose to study the field? What do econometricians do and what tools do they use to do it?
[Dr. Raviv] Very broadly, econometrics is the application of statistical methodologies to prove or disprove economic relations, mainly empirically, but also theoretically. As you can see, econometrics closely interacts with statistics. You can apply the same statistical techniques in medicine and in economics. The reason why when applying those techniques in medicine it is not dubbed “medicinometrics” is probably because economic models have some theoretical underpinnings that require some additional economic training.
What type of person might choose to study econometrics? Drawing from teaching others and from my own experience, unless you are really talented, you need to be able to sit for long hours, to almost “take it personally” when you can’t figure out something. Theoretical or practical debugging are an indispensable part of the process. Like everything else, it is easy to do when you enjoy it. So if you like economics or finance, you can sharpen your statistical skills and there you have it.
[OnlineEducation.com] Their goals and methods might vary, but one would think data analysts and scientists still use many of the same skills and tools. Can you share some of them? In your opinion, what skills, programming languages, or technologies are most important for future analytics professionals to learn?
[Dr. Raviv] The following are helpful to know:
- Ability to read basic math. When I grew up, I disliked [math] very much. A bachelor in Economics, which is a soft science, did not prepare me well for leaping into Statistics, which is an exact science. There is simply no getting around it. No one is going to bother writing half an A4 page for you if they can explain it with a single equation.
- Programming. Modern statistics is aided by computers. Advanced methods for both prediction and inference require what is called ‘sampling.’ You can’t do that without some basic programming knowledge. My own preference is for the open source R software. I know many who are happy with C, Python and MATLAB as well.
- Data visualization and communication. Programming will help you visualize your data, which is very important. It is important so that you can easily track what you actually do, and it is very important for communication. There is much to know about visualization. Properly presenting your analysis goes a long way. A well-cited book is that by Tufte, The Visual Display of Quantitative Information. You need to be clear about what is it that you are doing. If it is a ‘black-box’ method that you are using, then you need to be clear about that as well.
Finally, looking maybe a decade ahead, I would encourage newcomers to learn how to perform parallel computing. There are many good ideas out there that require enormous computing power, combine that with the fact that it is much easier to learn how to parallel today than it used to be, and this advice is a no-brainer.
[OnlineEducation.com] While researching analytics careers can give one valuable perspective, knowing and doing are not the same thing. Drawing from your experience, what is it really like to be a quantitative financial strategist? Can you describe a typical day, including the skills you use, tasks you perform, and tools you need to do them?
[Dr. Raviv] I will say three things:
In financial markets, you are often busy not with forecasting, but with signal(s) processing. At each point in time, there is a lot of movement, and a lot of it is noise. We try to distill the important bits from the ever large pool of information. We try to disentangle the signal from that noise. The way to go about it is by monitoring large amounts of economic data. Part of my work is bringing this data so that it can be quickly digested. Filtering what is important, aggregating and summarizing it. I often use volatility estimation, correlation modeling and bootstrapping. It is important to be able to program, to visualize, and to communicate the findings with clarity.
Another part of the work is handling more specific questions that require particular attention. Let me give a concrete example. There has been a sharp drop in oil prices recently. Like everything else, the price of oil is driven by supply and demand forces. It is hard to pinpoint how much of that drop has been due to supply and much was driven by demand. The answer carries meaningful investment implications. We spend a lot of time digging into this. Simple regressions simply won’t suffice. We scanned other existing research. We found a couple of relevant papers, one published in American Economic Review and another in the Journal of Applied Econometrics. You need to know what other relevant research is out there; to be able to read and understand it; to program it yourself so that you have control over it (as opposed to reporting what someone else did), making sure it is indeed a robust analysis; and to be able to adjust it to your own needs and maybe even improve on it.
Finally, and importantly, you have to be able to communicate the analysis and results to different audiences. Especially in investments, “Take my word for it,” is not a sustainable approach.