Powered By LEAN: Ternary Intelligence
In the next edition of our Powered By LEAN blog series, we feature Pavel Paramonov, Ph.D., Chief Science and Technology Officer at computational finance company Ternary Intelligence. Pavel recently spoke with our CEO Jared Broad to help us learn more about his company and perspective on the market, how LEAN helps with backtesting and strategy development, and much more!
Broad: Who is Ternary Intelligence and what is your role at the company? What’s unique about your business?
Paramonov: Ternary Intelligence is a computational finance company with a proprietary market instability detection technology rooted in statistical physics and complex systems science. Based in San Francisco, we provide signal subscription services to hedge funds and now also offer a retail investment product Hedged Bitcoin, available globally through the Uphold platform.
As a co-founder and Chief Science and Technology Officer at Ternary Intelligence, I have developed our proprietary market analysis methodology and brought the implementation into production with the associated cloud infrastructure.
Experienced hedge fund quants consider our multi-disciplinary approach as something that offers a fresh perspective compared to more traditional, classical and typically available finance tools. We analyze the dynamics of financial markets with the same science-based methodology regardless of the asset class, and that provides an attractive alternative to a wide range of predominantly data-driven signals that constantly need to be tuned.
With respect to our Hedged Bitcoin offering, what sets it apart is the fact that it makes a hedge fund-grade proprietary technology available globally for any retail investor, while providing a fully-algorithmic exposure to the cryptocurrency ecosystem with a more tolerable risk.
B: How did you get your start in the industry and what led you to Ternary Intelligence?
P: Looking at abrupt transitions in complex interacting systems within the framework of statistical physics has always been within my toolset, but it was not until 2013-2014 that I became interested in how financial markets are amenable to such analyses. A particular push in that direction came from the Bitcoin market that was roaring in the media at that time. While fascinating from a computer science and cryptography perspective, Bitcoin attracted my interest first and foremost as a global experiment in the dynamics of investor crowding with the observables that were easily available to track. My business partner Beau Giannini recognized that besides playing with the cryptocurrency herding, the approach should also be applied to more capitalized financial assets, and that gave birth to Ternary Intelligence.
B: What is the Ternary Intelligence perspective on the markets?
P: We emphasize the fact that similar underlying behavioral finance drivers govern investor crowding in a variety of markets. We treat markets as interacting systems of units (traders, funds) that have a limited set of options at each point in time: to buy, sell, or not do anything (a “ternary” set of options, to which the company owes its business name). At this level of abstraction, the toolbox of computational physics can help identify specific points in time when the market dynamics become partially predictable.
Most science-based approaches are not created out of thin air and our case is not an exception: the methodology was inspired by academic works in the field of econophysics treating financial markets in a similar way. We have developed a number of unique elements on top of that foundation and brought the resulting product to the financial industry.
B: We see that Ternary Intelligence creates investment signals by integrating theoretical physics with computational finance, in addition to pattern recognition. Can you share more about your investment style?
P: Coming from the applied physics background, I see beauty in universal patterns. One of the most attractive features of our approach is its universality: we apply exactly the same diagnostics procedure to any publicly traded asset class. This works top-down with global macro indicators as well as bottom-up with single equities, all across multiple time scales. Our market instability signals constitute a research product – they can be combined with various investment styles and so far we have had integration endeavors with long/short equities, commodity strategies, volatility trading, and the crypto market.
One of the most challenging aspects of applying our methodology is identifying specific patterns that we expect to see in market prices from a theoretical perspective. Since the underlying physics is fairly universal, we needed a pattern recognition approach that would also be agnostic to specific asset classes and would not require time-varying tuning parameters.
To that end, we have developed a proprietary methodology that is consistent and universal in a sense that pattern detection is done the same way for any publicly traded asset, be that an equity, equity index, commodity, or crypto. There is no tuning/tweaking over the time history involved, which is important for being prepared for new market turns that have never been seen before data-wise.
B: What made you chose LEAN for your trading engine? What were some previous pain points you experienced with prior backtesting/trading platforms?
P: I have been watching LEAN evolving from its early days, first hearing about the QuantConnect platform concept around 2012 when I was involved in a Biotechnology startup and we were part of the Startup Chile program which QuantConnect participated in as well. Later on, when Ternary Intelligence was formed, I kept watching LEAN more closely as a candidate for our research and execution engine.
One very common problem with various platforms was the disconnect between the backtesting and live execution code. Subtle bugs could introduce forward bias in backtesting and compromise the reality checks for trading strategies. Another issue we frequently faced was the flexibility requirements in accommodating custom data inputs. This was a particularly important point for us since we generate signals at multiple time scales with specific numerical characteristics for each scale, and need all those custom inputs to be robustly integrated into the strategies.
Striving to accommodate our custom inputs as well as prevent any forward bias in backtesting, I ended up writing a separate platform in a reactive programming style that powered our smaller-scale cryptocurrency experiments back in 2014. However, I quickly found that maintaining the integration with evolving crypto exchanges frequently consumed more time than the strategy development itself. As LEAN matured and QuantConnect extended its data library to include options, futures, and crypto, it gradually became our preferred research and production environment.
B: How does an open-source approach benefit your business? What are the dynamics that you are focused on?
P: Using open source components enabled a faster strategy development cycle, more straightforward market data ingestion, as well as more polished trade execution. I believe this approach may greatly benefit large institutions as well – it helps to have lots of eyes watching the same code base of backtesting and execution engines, and that is what we are focused on in the open source ecosystem.
B: How can LEAN impact the future? What sets LEAN apart from other solutions?
P: To me the most important points that set LEAN and QuantConnect apart from other platforms are (i) fully open-sourced code that anyone can run, analyze, and tinker with independently of the QuantConnect infrastructure, (ii) consistency between the backtesting and the execution code, and (iii) QuantConnect’s extensive data library covering not only equities but also options, futures, and crypto. As far as impacting the future, QuantConnect’s slogan of democratizing finance does seem to be coming closer to reality these days.
B: What’s next in the world of investment management?
P: In brief, I would expect two key trends: crowdsourcing and asset class diversification.
Crowdsourcing of investment strategies is already a term that starts to resonate with some quantitative funds, and platforms such as QuantConnect play a central role in that. Smaller-scale investment managers that do not have in-house quant teams are also likely to realize the advantages of strategy crowdsourcing and to start integrating that into their practices.
Most asset managers in large family offices and hedge funds are still restricting their portfolios within the scope of “traditional” asset classes. However, diversification in the asset types is already seen with some smaller-size managers that already invest in cryptocurrency. We believe this trend is likely to continue.
B: Anything else you’d like to add about your experience with QuantConnect?
P: I am still amazed by the amount of work and infrastructure resources that the QuantConnect team deployed to make their data library available for the users. At some point, I evaluated the scope of work to assemble our own in-house minute-resolution options data library and concluded we would not proceed with that undertaking in a small business setting. QuantConnect did the hard work and made it possible to run backtests with that type of data in a fairly straightforward manner, which I believe is a major achievement.
More recent efforts at QuantConnect related to the Algorithm Framework also seem to help the community by emphasizing systemic thinking about trading strategy elements, which is how the workflow is organized in better quant fund team.