Interview with Morgan Slade, CEO of CloudQuant
With over 20 years of experience as a trader, portfolio manager, executive, and entrepreneur, Morgan Slade is now the CEO of CloudQuant, a cloud based quantitative strategy incubator and systematic investment fund. He has built quantitative trading businesses at some of the world’s largest hedge funds and Investment Banks over the last 20 years. Previously he was Global Head of Equity Trading at Allston Trading, a major HFT firm, Portfolio Manager in HFT at Citadel, and head of Global Systematic Trading for Kershner Trading Group. Morgan earned a BS and MS in Materials Science and Engineering with a concentration in Finance from MIT.
At CloudQuant, the company uses machine learning, alternative datasets, cloud computing, and crowdsourcing to generate alpha at scale. You can follow Morgan and learn more about CloudQuant by connecting via twitter @jmorganslade or LinkedIn https://www.linkedin.com/in/morganslade/. You can learn more about the CloudQuant team here: https://info.cloudquant.com/team/ or by following them via twitter @CloudQuant.
Grace: You started your career as an engineer and you’ve been in the industry as a quant trader for quite some time. How did you pivot your career and what inspired you to get into Quant trading?
When I was in MIT, there was a well-worn path from MIT to the Chicago and New York trading floors. I did a five year master’s program and spent some time at Motorola and Intel. As part of that I spent some time away from school working on my master’s thesis for Intel developing a sheer strain measurement technique for finite element modeling and finding ways to validate mathematical models using empirical data. When I returned to school after having spent some time in industry I started to talk to traders and I realized it sounded like their lives were a lot more exciting than an engineer for a Fortune 500 company. The excitement, the dynamism, and the need to keep learning new things was what drew me to that. I found an opportunity in a boutique investment firm and then transitioned into a positon at Millburn-Ridgefield which was one of the oldest CTAs at this point. They were looking for quantitative traders and I had a really unique opportunity to start trading right off the bat at a large multibillion dollar futures hedge fund and I had a chance to not only trade during New York hours but I also ended up covering the book in the Far East as well as in Europe for various parts of my career there. I actually worked a second shift in the research department team developing trading strategies so I was putting in long hours, sometimes sixteen hour days; trading for eight and then doing research for another eight. It was a pretty intense time in my life and I was able to kind of get a good handle on market impact of large trades in execution quality and the importance of information flow. We traded futures on the exchanges and had direct lines to most of the exchanges and had brokers on the floor that would pick up and work for us directly but we also traded cash FX over the phone with all the major dealers.
Grace: Going back to what you are doing now with CloudQuant, I understand that you have a trading strategy incubator where your team has the experience, the technology, the capital and allows algo traders to essentially get a strategy funded. Can you tell us more about that and the vision behind it?
We started building CloudQuant in 2011 and we recognized that there were inefficiencies in the market and hundreds of traders were looking for these [inefficiencies] but many of them did not have the solid skillset to search for alpha opportunities in a scalable way. We did have a core team of quants that continued to use CloudQuant internally to run a quantitative fund and we had good success with that. As an extension, the reason we built this trading and research ecosystem was ultimately to offer to crowd researchers so that we could level the playing field so that engineers or data scientists or professionals anywhere in the world would have the ability to gain access to institutional quality research tools. Now we think that the trade execution ironically, although it’s made with algorithms, is something that’s very difficult to understand for people who are not native to the financial industry. So we actually draw the line at taking care of the operation of the trade and the compliance and risk management ourselves and we simply leave the strategies to crowd researchers.
We are taking those four things [machine learning, alternative datasets, cloud computing, and crowdsourcing] that we bring together as ingredients to this very special recipe that is powerful and will ultimately allow us to generate alpha at scale for the first time.
About twelve months ago we set up our website and created a front end for anybody with internet connection to be able to start developing tick based trading strategies using microsecond level trade and quote data in the U.S. markets. We’ve seen tremendous success and appetite for these tools. Because we provide funding ourselves and work hand in glove with them not just to maximize our returns but also maximize their returns through a payout structure that compensates them for net trading profits from the strategy, we’re able to entice some very talented people to work with us and solve really difficult problems. So that’s the idea; we are taking advantage of machine learning, alternative datasets, cloud computing, and crowdsourcing. We are taking those four things that we bring together as ingredients to this very special recipe that is powerful and will ultimately allow us to generate alpha at scale for the first time. When I worked for Citadel, Allston Trading, Merrill Lynch and other places we only had a fixed number of people on our research team. It could be five people or it could be fifty people depending on which firm you’re at but at the end of the day it’s a very small group of people and they can only do so much research. What is exciting about CloudQuant is the possibilities of tapping into the twenty five million data scientists in the world or the millions of engineers that there are in the world that would love to have a chance to work on Wall Street but have never been given the opportunity.
Grace: I know there’s a lot of open source tools out there and you mentioned institutional quality tools available for backtesting, data visualization, and more. What are your thoughts on them? Are there any particular tools you use yourself or you prefer?
Morgan: We actually are developed using the anaconda stack of Python tools and we’ve actually partnered with Anaconda which is the world’s most popular Python stack for data science. Part of our development team came from Anaconda and worked with us over the last four or five years to create a scalable framework that allows us to support all of these researchers. That’s definitely a technology that we have leveraged and we use a lot of machine learning which is mostly open sourced. We have lot of active R&D projects internally having to do with artificial intelligence and deep learning. We have lots of alternative datasets that are evaluated for the crowd researchers to be able to utilize and those are typically very expensive and accessible if you worked at the top hedge fund. Because we have economies of scale and a technology platform that allows us to test them out without having the crowd researchers be actual licensed users and because of our proprietary technology we have the ability to direct thousands of users to look at particular datasets and have them identify sources of alpha that we normally would never get to. We just wouldn’t have the time in the day to look at them. Ultimately that will give us the biggest advantages of having more talented people at the keyboards looking for patterns, looking for ideas, and contributing their unique ideas and perspective.
Grace: You mentioned that you focus on machine learning as well as open source. When you first started getting into machine learning what were some of the challenges you faced and how did you overcome those hurdles?
Morgan: I started machine learning in ’99 before it was famous. It became my go to approach probably by ’00-’01 and was my preferred way of solving many of the problems that we try to solve in quantitative investing. Some of the challenges are obviously a tradeoff between getting a more accurate model and curve fitting. There’s a lot of rules you can follow to try and prevent that but each type of machine learning algorithm is different. In general, for any free parameters in the strategy that I have I’d like to see about a hundred different independent trading decisions being made before I have any confidence that I’ve not curve fit. In general, I like to look for things that have a good Sharpe ratio, a very high rolling twenty day win rate and a very consistent rolling ninety day win rate over time. That usually gives you a pretty good sense for whether things are going to be able to withstand drawdowns before you get out of them and give up. It also gives you some very interesting data to look at as you’re trading to keep things in perspective and you can know what to expect. I also look at something that I call the Kelly edge. Basically if you look at the odds ratio of the average profit to the average loss you can imply a fair value win rate. It’s really the rate at which you need to win in order to breakeven over the long run and if I take that fair value win rate or a strategy that’s paying out one to one odds, my gains are the same size as my losers. Then the fair value win rate of that is 50% so it’s easy to understand if you win 52%, fifty two percent of the time then you have two percent Kelly Edge. Most strategies don’t have that kind of payout, they have some type of skewness to their payout and so it could be one to three, it could be one to four or five. At any rate you have some fair value payoff for every strategy on average and so we try and quantify the statistical edge you have as being your win rate above and beyond the fair value win rate for your strategy.
Grace: What are some of the come pitfalls of backtesting and what advice do you have for newer traders to avoid overfitting?
Keeping things simpler is better than just throwing in the kitchen sink. When I was first taught to develop strategies I was really struggling with how to pinpoint what was going wrong when something didn’t work and I always faulted myself because maybe I’m not thinking about this clearly or something. When we set out to trade strategies there really is a better methodology that we can teach people. Do the prediction first, and build a workflow for your research process that really focuses on maximizing your prediction quality first.
When we set out to trade strategies there really is a better methodology that we can teach people
Then once you get to good prediction quality move on to the next step where you start allocating risk and deciding how large of a trade to send and if you are trading a basket how to allocate risk amongst the various things in your portfolio. Finally when it comes to trade execution really analyze the trade execution part of it separately. Look at your arrival price and look at your slippage and look at that as almost a separate problem. By breaking it up into three pieces it’s much more transparent where things are going wrong and I’ve found that people are much more efficient at producing workable strategies when they follow this simple three step approach.
Grace: What are some other ways of prolonging the viability of a trading system?
Morgan: With quantitative traders and researchers and data scientists their food is data and so by being very efficient and feeding your people new datasets that they can sink their teeth into, that’s ultimately how we help them bridge the disconnect between information and price action. If we can predict price action with information that’s better and with techniques that are smarter and with insights that are deeper then we will always be one step ahead of our competition. In the end the world continues to evolve and the markets become more efficient. We’re just hoping to create a way to be more efficient than anyone else at managing investor’s capital.
Kurt: Do you think it is possible for someone from a programming background or an engineering background to successfully build a system and implement it into the market with no knowledge of trading?
Morgan: Well I’m skeptical that they could implement it with no knowledge of trading but that’s exactly why we are set up the way we are. We don’t expect them to know anything about what we call trade expression. We just want them to help us predict prices and then we will lease the strategy from them and take it from there. We have experts who have been in the industry doing high frequency market making for a long time who inform our trade expression platform. We leverage that expertise to ensure that we’re getting the best execution we can given the strategy and we don’t really want or expect crowd researchers to become experts in the U.S. market microstructure. It’s very complicated and it could be a full time job in and of itself and they’re bound to not be doing something optimally. So we have trained professionals who work just for the crowd researchers and their only job is to make sure that they’re making as much money as possible for the crowd researchers and the book itself.
Grace: In regards to datasets, do you think an algorithmic trader inputting only open, high, low, close, and volume in their model can still have an edge in the market?
Morgan: I would say combined with another alpha signal or two sure. There’s definitely a lot of validity to those technical strategies which is why they’ve been around for so long. But as those trades become crowded because it’s fairly simple to build models like that what you see is people need some additional overlay or check to make sure that it’s indeed a good signal and it’s going to follow through. So that’s where I think these additional datasets come in to provide some additional information that helps you identify whether something is sustainable or not. We see a lot of older strategies that have decayed over time be rejuvenated with the addition of alternative datasets.
We see a lot of older strategies that have decayed over time be rejuvenated with the addition of alternative datasets.
Grace: In your experience, and bringing it back to the retail trader, what is the most common struggle that a new trader will go through when they start algo trading?
Morgan: I think that curve fitting is obviously a big problem and just lacking the perspective. If you come up with your own you’re obviously going to try and select the very best one you can come up with. Inevitably that’s the one that’s going to curve fit. New traders struggle with the tension between wanting to make more money and wanting to be intellectually honest with themselves about whether they have done too much data snooping or not. There’s a temptation to go to the dark side there and we talk about it internally all the time. Even people who do it for a living and get paid very well to do it are tempted by the dark side of curve fitting and they have to actively protect against it.
Grace: What resources or books do you recommend for someone who is looking to get into algo trading?
Morgan: One that’s recently come out that’s interesting is called Algorithmic Trading: Winning Strategies and Their Rationale by Ernest Chan. There’s a book called Investing by the Numbers by Jarrod W. Wilcox that I read and found quite useful earlier in my career. It has an overview of a couple different strategies and a couple different methodologies for pricing equities and building portfolios of equities. A fun book that may appeal to people is called You can be a Stock Market Genius by Joel Greenblatt. That is for people who are willing to roll up their sleeves and dive into the financials, read all the EDGAR filings, read the prospectus and understand capital structure which are things that are hard to automate. Those are interesting books to read and give you a good perspective on how the markets work.
Click here to watch an Intro to Machine Learning by Trevor Trinkino of CloudQuant.
Risk Warning: The FXCM Group does not guarantee accuracy and will not accept liability for any loss or damage which arise directly or indirectly from use of or reliance on information contained within the webinars. The FXCM Group may provide general commentary which is not intended as investment advice and must not be construed as such. FX/CFD trading carries a risk of losses in excess of your deposited funds and may not be suitable for all investors. Please ensure that you fully understand the risks involved.