ML Quantics interview 5 3 16

hello everybody I'm alexic rubber off i'ma chief scientist at nitro and today we host the meetup by san francisco by air a machine learning group which is a partner made up of many other meetups because here including its of scholars of sparks a text we have a tradition of hosting this great metal and today is the first time we have fintech talk we have guys from quantity x martin in virginia and they're gonna introduce themselves and mention what they will be talking about today okay thank you yeah I'm margin the CEO and founder of quantity X quantity X is the first marketplace for trading algorithms so we connect user-generated quantitative trading systems trading algorithms with capital from institutional investors and our users participate in the profits of their an algorithm without investing their own money we provide our users with everything they need to build trading algorithms which is primarily software and data so we provide them with free and open source software in matlab and python to build their trading algorithms and 25 years of historical market data and yeah i'm very happy to be here tonight with jenya one of our best quants you want our q3 competition and we're happy and proud the trade is algorithm with a million-dollar in futures right now jenya yes i'm jenny was going off and i am an idealized client and user of the contacts platform because i actually want but some of my friends also once so it's not that uncommon to win with their platform i'm a grad student at Cal Tech and my primary job is doing science and this is what I'm going to do for the future and finance is just something that supports my passion for sense this sounds really exciting so so basically you guys really harnessing multiple things happening in the internet you have the basically supply of data you have a you you supply for some kind of crowdsourcing so can you tell us what the history is how did this come together okay the history of contacts started in Europe I worked as a quant in in the industry for almost ten years I've built up a quantitative research firm in Zurich and I did a lot of recruiting and training of novice Kwan's and that's how I realized that the industry today wastes a lot of its potential from we were a small quantitative research firm but even we got 100 applications / quant that we hired so there are 99 others out there who eventually don't make it into quantitative finance and yeah they are still capable and we want to empower them with quantity X so that's how I got the idea to found this company to empower the ninety-nine dollars who don't make it into quantitative finance either say how did you find any yeah well we started quantity X with through hosting competitions so we wanted to incentivize primarily students on good US schools like Stanford and Kaltag and Berkeley to participate in our competitions and to submit trading algorithms to work as a quant even though they have never done that before yeah and jenya was one of our most successful participants and he learned about it on on Cal Tech but maybe he can tell more about that he of course So Cal Tech has quite a interesting kind of story of victories with coin checks because I actually learned about it because a culture graduate student a few years older than me was able to win the previous competition and that actually sounded really amazing at the time so I contacted that person and was reassured that this is not a scam of any sort and the there's actual money that the person who received after her algorithm been traded and so eventually I realized that okay maybe I should also give it a try and then like it kind of went on as a chain reaction so other people learned about my success and the actually one of the one of the students that worked with me started trading and what many of the people in CS started trading with context so started writing algorithms for that platform and we share ideas it's really amazing and but even even if in words we have the same idea when we actually go and write algorithms they have like directly opposite performances so the implementation is really key and this is what we keep to ourselves mmm well ask maybe of to follow up so what kind of knowledge can you tell us first of all about the science we can premiere area and what kind of knowledge do you need to write trading algorithms oh I guess that that problem doesn't look like machine learning at first but I actually got excited about it exactly because I understood that it is similar to other machine learning problems so I can approach it with that same tool box even without knowing anything about finance so this is what's great about contacts is that the barrier of entry is really low a person can barely know like two or three finance jargon words and already make it through the tool box and use it so on data hand if you actually do know something about the technical and fundamental tracing or you have had experience as an intern say in a hedge fund or a finance company or a bank that experience may also help you to write algorithms just those will be those will be a different kind of algorithms then a machine learning person all right so I will ask you I mean a lot of traders are high frequency traders and like there is a lot of books and media you know press about how they're fighting for every millisecond latency so are you in that space can you tell me about the kind of how do you compare to these guys how can you compete against the single giant and it is yeah we're absolutely not in the high frequency trading space we work with relatively low frequency data in our case it's end of date data and we work with futures these are the most scalable most liquid markets worldwide and they have the nice property to cover a universe of assets that is very uncorrelated so futures of crude oil of Japanese yen of wheat and so on so many diverse instruments and we work with low frequency data for reasons of scalability of the strategies we want our institutional investors to be able to manage multiple millions every strategy on our platform that's why we need markets that can handle such a volume and the second reason is since we are a crowd source platform we also have to protect of course our institutional investors against any malicious practices that could be applied on from people who actually know the code of the algorithms that we trade so we have the problem that somebody develops an algorithm and we trade it and iterated for an institutional investor with a lot of capital so somebody could use that knowledge and arbitrage or front-run the investor all right so in order to protect the investors from that we also work with daily market data because this way we have a certain discretion over when we actually execute the trade so we get a signal from from trading algorithm but when within the session we execute it is unknown to the Quan to develop the strategy and this way we can protect the investors got it this is very very interesting how how do you convince institutional investors that they can trust a bottle we give them much more information than quantitative hedge funds normally would so more information we never disclose the the secret sauce of the algorithm but no quantitative hedge fund would do that so that remains the intellectual property of the quant but we give them much more information and I think the best way to build up trust is that we trust the algorithms enough to trade it with our own capital which is what we're doing so we trade for example Chinese algorithm with a million yeah and that shows institutional investors that it actually works so this way we build up live track record we show them that the simulation matches the life results and that it actually works servania I we'll kind of follow up on this amount like Mark mentioned your turn the million dollars worth throw algorithm how do you feel about this I mean how did you progress did you build up the first are trading in thousands what was the progression oh well I am certainly not feeling like a millionaire hey I guess it's it's like a fraction of a fraction of a pipe because the algorithm has to win and then I get the fraction of what it one so when it comes down to me it's actually not that much however it's still comparable with what I get from any other part-time job which is amazing so I i guess the actual algorithms are kind of the way we the competition is scale-invariant so it doesn't really matter what is the total sum of money that you trade with the winner is still the same so for the competition I just developed an algorithm that completely is completely ignorant about those issues of scalability of the algorithm and whether like the liquidity of of the markets will kind of bound it or not so my algorithm was like from the start about a million dollars or like any other amount mm-hmm interesting interesting so and and kind of wood is I would say lets you know I think it's super exciting to you know our members and our viewers what kind of wood is the path to try this like you know your graduation contact that's a pretty high bar what kind of the basic like what kind of knowledge is most useful so much' learning like what aspects of machine learning are most important for somebody who wants to try this kind of stuff from your experience well I I will cover it more in in the following talk which is actually on youtube with contacts however if I need to sum it up and like really short sentence then i am not a machine learning major at Caltech right so i was able to do it just by knowing very basic ideas about machine learning that you can pick up really from everywhere as soon as you start trying the job for those data science positions you will get some kind of machine learning technical assignment what I socialize in the local tech I'm a theoretical physicist and I actually work in quantum information which then kind of connected with my hot babe because in quantum information we also build tools that potentially will improve the machine learning practices but so far I i can say that i did any quantum machine learning even though like for fun maybe somebody named their algorithm this way so yeah I guess any kind of statistics on machine learning experience for working with the data experience would be good enough for me this experience came mostly from a computer vision class that I took at Caltech I never actually took a machine learning class mm-hmm so I was interested in the texting pedestrians on the streets but then it turned out that after I learned that finance wasn't actually that hard then I started detecting a huge profits in the markets this is really really interesting yeah and this is really cool so we're looking forward to your talk thanks for coming this is exciting and I know that Martin's speaking on this subject that data by the by conference which comes in the middle of my where we'll cover it more and we're looking forward to have more FinTech talks at our meetups thanks