Bringing Traditional Financial Influence Into Crypto Markets With Dfi Labs
Dfi Labs is a data and tech company as well as a cryptocurrency asset manager. Olivier Chevillon, Founder at Dfi Labs, joined Paul Gordon, editor at Coinscrum Markets, to discuss the parallels between traditional finance and the cryptocurrency industry as well as the role that the company plays in the market.
As a traditional finance guy, with two decades of experience on the trading and quant side, Olivier has always been passionate about innovation. In crypto, he and his team found something that matched their interests with a lot of data and opportunities through which they could apply everything they’ve learned and try to shape something new and exciting.
When Olivier first heard about bitcoin in 2012 or so, his first reaction was that it was a scam and he was determined to stay as far away from it as possible. He says now that it was because he didn’t take the time to understand it.
(6:46) “When you are in traditional finance, you are used to things having proven themselves throughout time. And everything which is too good to be true is too good to be true for a reason. And so then the question is, is this because of a lack of liquidity that it’s too good to be true? Or is it just because it’s a scam?…I didn’t have time to dig into it, and so by default, I guess I took the shortcut of saying this is a scam,” said Olivier.
He was, however, a “massive fan” of blockchain, which led him to code in blockchain language Solidity and in turn allowed him to discover a lot of the data that was available on the blockchain. He was already drowning in data in his day job, trying to predict what will happen, and the blockchain provided even more data on something in which he had a ton of interest. So he thought, “Why not try to make it happen in this space.” That was 3.5 years ago, and today he is the founder of Dfi Labs.
Prior to crypto, he was creating strategies based on data, using machine learning with a view of setting up a stable pipeline to generate alpha on a fairly constant basis. In fact, that’s what he found so interesting about the cryptocurrency space — because there’s so much data.
(9:25) “There is a lot of value that you can extract from a background in traditional finance because it saves you a lot of errors. And in trading, my first boss always told me, ‘it’s not about making money, it’s about avoiding losing money.’ And if you’re avoiding many traps, then definitely in crypto, you’re very close to being a very good product,” said Olivier.
The crypto universe, however, needs to step up to the standards of traditional finance, he added.
Clean and Historical Market Data
In the traditional markets, there is what’s known as traditional data, where the data is very clean, Olivier explained. In the cryptocurrency space, however, you can take 10 different exchanges and you don’t exactly know what the liquidity is, the order books are “ci, comme ça,” he said, adding that the quality of data is much lower. It’s neither unified nor clean.
Secondly, there is a lack of clarity surrounding the length of data that is available and how far you should be looking at. Did it start in 2009? 2012? 2015-2016 or 2018-2019? The question of how long your time series are is also very different from the traditional space, where the length of the data that you’re looking at and the quality are much higher.
(12:21) “ So the number of assumptions that you have to make as to whose data is much lower, which means that when you’re modelizing, it makes it a lot easier on the traditional space. Here you have a lot more assumption that you need to clearly state when you’re doing it. So I think in terms of the data which are available, it’s very different,” said Olivier.
The infrastructure landscape is also very difficult in the cryptocurrency spacel, as there are not the same options that exist in the equity space. Even on exchanges like Binance, you buy the offer, sell the bid or just stay on the bid. It’s very immature in terms of the offering, he said, adding that if you look at the number of scams that have occurred involving the security of data and the keys, you’re at levels that are not acceptable to traditional investors who are going to put a lot of money at risk.
(13:48) “So everywhere there’s a couple of key issues for the whole industry to solve before being able to be fully mainstream to all types of investors,” said Olivier.
Dfi Labs initially started by taking the full sample, using a cut in 2017. But they didn’t believe this was very accurate because it led to very skewed data in which there were a lot of very positive results and very few to the contrary on some other cryptocurrencies. So they moved from using the full sample to using a certain number of holding windows.
They use different lengths of memory on which they train their model and then stack all those models, or add them together so that they have less dependency on one single point. It results in 20 people voting together as to what they believe is the right outcome. Basically, They are using holding windows and they are stacking them so that it’s a vote between a certain number of models that are alternating on different holding windows.
It’s how they’ve chosen to do things because these data sets better reflect the investors that are in the crypto space, where most of them have not been here for 15 years or even five years. They’ve been here for one to three years. Dfi Labs is trying to extract how they think .
Copy & Pasting vs. Customization
When Olivier was on the equity side, he dealt with cross-sectional trading, so buying a basket and selling a basket with the goal of achieving alpha, or the outperformance of one basket vs. the other. Basically it was about trying to guess what the trend will be for the next day.
In the cryptocurrency space, you’re very much trying to predict is it going up or is it going down, but for one basket of underlying, which is a very different way of looking at things.
(18:50); “I would say what I’ve done is really, I’ve brought modelization that I was using. So for instance I’ve described how we’re using holding windows on which every single holding window gets a model that we train and then makes a prediction. Then you have 10 models which all make a prediction…So this is one thing that I brought,” said Olivier.
The other thing he focuses on is risk, which he says is the most important thing in the crypto asset management space, where you can make a bundle but you can also lose a bundle, which isn’t good for you or the industry.
(19:35) “And so being focused on risk more than on returns because you will get them if you get the risk right, is another key component. So understanding that it’s not about maximizing returns but it’s about maximizing returns under constraints…So understanding risk, this is something also that I brought from my days,” said Olivier.
He also focuses a lot on the data and the quality of that data. In addition to momentum and reversal, which are key components in the decision-making process, there are plenty of other sources of data, including blockchain data, sentiment data, other financial assets, DeFi, etc.
(20:50) “There’s plenty of types of data that people should be looking at. Because in the equity space, you’re always looking at new sources of alpha, new types of data. This was something that was part of our DNA. So trying to extract data from different types of sources of data, combining them, devising smart pipelines under certain constraints, that’s what I’ve brought from my days in the equity space much more than a model that I was using and that I copy and pasted on the crypto side,” said Olivier.
Dfi Labs Talent
Dfi is based in Paris. While Olivier lived his entire professional life in the UK, he came back a couple of years ago. The great thing about Paris is there’s a lot of talent on the data science side, which is a focus of the company. And so he met his co-founder and they decided to work on this project. The team was two guys: one with a lot of finance experience and one guy “who was a freak in terms of data and machine learning, that’s how we started,” said Olivier.
They quickly learned that on the crypto side, you need to be able to trade 24/7, so they added a guy on the tech side and they became very robust. They had big amptions and decided to make their project into a company, adding someone else with experience as the COO of a crypto firm, which rounded out the core team. Olivier then tapped into his network of data scientists from his engineering school in France, which is the top one in the country for statistics.
(27:56) “And so it’s really like a data firm. We’ve got quite a few data scientists. And that’s the idea — creating a tech firm able to extract value from the crypto market. We are only focusing on cryptos,” he said.
Dfi Labs Strategy
Dfi Labs at its core is a data firm that’s able to extract value from data in the crypto space. They have a couple of crypto fund of funds that are interested in what they do in addition to high-net-worth individuals who also love it. Those are the company’s core clients currently. They are not aiming to distribute their product on their own. The idea, Olivier said, is very much to have people doing the distribution part and running everything for them, and the core team acting as an advisor.
Coming from the equity space, Olivier believes the MiCA regulation may be annoying but it is also great, saying:
(30:37) “We are an industry which is constrained by how young it is. But we are also constrained by the lack of [regulation]. If you’re like a proper investor, you want mature products…you want them to be bulletproof. And you want the set up to be bulletproof. As long as we don’t have this regulatory side to it, this market for investors will always be a dwarf,” said Olivier.
He views it very much as an opportunity, saying the AMF in France has been supportive to the whole industry.
Find out more about Dfi Labs at dfi-labs.com