BoGe Partners started as a research project in early 2020, diving deep into the rapidly evolving world of cryptocurrency trading. The journey began with Charlie and Pierre-Antoine, who had been successfully arbitraging digital asset prices across various cryptocurrency exchanges since 2016. Recognizing the potential of expanding their strategies, they welcomed Alexis to the team. With his strong background in quantitative techniques from the managed futures space, Alexis set out to develop market-making strategies tailored to the unique dynamics of digital assets.

The Birth of a Research Project

The purpose of quantitative research is to empirically falsify hypotheses about market behavior and then form trading strategies that transform these findings into a profit. Despite its inherent objectivity, it is susceptible to tacit biases that can sway the outcomes of experiments.

Consider this scenario: a researcher derives an idea from an article, conducts tests, and obtains statistically significant results. Are these results reliable? Not necessarily. If the author of the article tested numerous ideas and only published the most favorable one, then regardless of the researcher’s diligence, determining the solidity of the results becomes challenging.

These tacit biases are prevalent in the investment industry where everyone shares the same information, reads the same research articles, listens to the same podcasts, follows the same blogs, and attends the same conferences, resulting in a lack of independent or novel experiments.

Lacking prior experience in digital asset / cryptocurrency trading outside cross-exchange price arbitrage provided an opportunity to conduct a research experiment with minimal bias.

To ensure objectivity and avoid common pitfalls in traditional research, we developed an automated system using a genetic program (GP) to create trading agents. We defined the structure of these agents and populated a library with diverse trading rules. The GP autonomously generated and refined trading agents solely based on data.

The Anatomy of a Trading Agent

A trading agent (TA) consists of three primary components:

• An entry rule
• A sizing rule
• An exit rule

This flexible setup lets us create different types of TAs, such as those using rules within a single market or between markets, and rules that are either binary or continuous. We built a specialized simulation engine to evaluate the historical performance of these TAs.

At this stage, we ensured our research remained unbiased by external factors. However, like any optimization tool, the GP can introduce its own biases as it searches for better solutions.

Quantitative research cannot definitively prove the presence of predictive signals in data. However, by observing how the GP behaves when we eliminate potential predictive signals from the data, we can infer insights about their potential existence in actual data, such as quicker convergence to solutions or variations in the distribution of fitness values.

Due to the limited historical data of cryptocurrencies, we utilized intraday data to increase sample size. We split the data into two sets covering complete market cycles (bear and bull): in-sample (IS) data for GP training and TA portfolio construction, and out-of-sample (OOS) data for validating the final model.

Building a Robust Trading Model

The GP has identified numerous promising trading agent candidates that we combined into an ensemble portfolio using a stepwise portfolio construction method, focusing on achieving robust internal diversification among TAs. We then finalized the trading model by adding overlays to manage variables like exposure, risk, and turnover.

Analyzing the Results

We simulated the trading model on the in-sample data set to form expectations about its risk-adjusted performance. Then prior to simulating it on the out-of-sample data set, we considered three potential outcomes:

• If OOS results mirrored IS results, we would proceed to live implementation.
• If OOS results showed slight degradation but met industry standards, we would conduct further research to refine the model.
• If OOS results significantly underperformed IS results, we would halt the project.
• With OOS results slightly outperforming IS results, we proceeded to develop the execution algorithm and launched the fully automated trading program in early May 2020. By July 2020, the program was already seeded with several hundred thousands of USD of our own money.

FROM RESEARCH TO Live trading

Since its inception in May 2020, the trading program has consistently delivered risk-adjusted performance in line with our initial research findings, both for in-sample and out-of-sample data sets. Our live trading experience now extends 50% longer than our initial research dataset.

While the genetic programming approach was crucial for initial discovery and the core logic of our models has remained unchanged, our trading program has undergone substantial evolution as we gained experience and learned from market feedback. We have implemented new models, refined existing ones, adjusted our instrument selection, significantly increased our trading capital from hundreds of thousands to millions of USD, and enhanced our execution algorithm to reduce costs and tracking errors, among other improvements.

We deepened our understanding through actively trading these models amid the highly volatile bull and bear markets that cryptocurrencies have witnessed since 2020.

What’s next for BOGE PARTNERS?

Arnaud joined our team to contribute his extensive experience in administration and sales. In mid-2024, we founded BoGe Partners and our mission is to provide innovative investment solutions to investors, particularly in Europe.

We provide qualified investors access to this trading strategy in an actively managed certificate (AMC) known as the BoGe Digital Assets AMC. For more information, please contact us.

We are also developing exciting new products, including the BoGe Diversified Assets AMC. This asset allocation model invests in global equities, global bonds, precious metals, managed futures, and digital assets. By employing a capital-efficient approach that uses both cash and derivative products, it targets a higher volatility level than is achievable with cash-only products, offering investors more exposure and potentially higher returns on their capital.