(Upcoming) Automated index management through ML

Elastic Net Regression in Crypto Fund Management: A Path to Outperformance

In the ever-evolving landscape of cryptocurrency markets, conventional investment strategies often fall short due to their static nature. Nex has been built with the philosophical saying "The only thing I know is that I know nothing" as a foundational principle of its values.

The utilization of cutting-edge machine learning techniques, particularly elastic net regression, is driven by our commitment to this principle, enabling continual and dynamic adaptation to market changes.

Our method is rooted in scientific rigor, notably through the use of the scikit-learn library and a strict policy to prevent look-ahead biases, ensuring that our strategies are both robust and forward-looking.

Scientific Approach and Look-Ahead Bias

Look-ahead bias occurs when a model includes data in its analysis that would not have been available at the time of trading, thereby skewing results and giving a false impression of predictive accuracy. To circumvent this, data up to the point of each backtest is exclusively utilized by our models, accurately mimicking real-world trading scenarios. This scientific integrity is crucial for developing models that genuinely perform well in live markets, not just on paper.

Furthermore, the dataset is randomly split using A/B testing methods, facilitating the assessment of the model's performance across different data segments independently. This approach not only simulates varied market conditions but also mitigates the risk of overfitting to a specific dataset structure. Additionally, the results of multiple runs are averaged to ensure consistency and reliability in our predictions.

Backtesting and Continuous Learning

Our backtesting results, derived from historical data, demonstrate the efficacy of our models under various market conditions. By updating our models on trailing historical data on a monthly base, the evolution of our strategies in response to new market dynamics is ensured, embodying a truly adaptive investment approach.

Two-Stage Selection Process

In the first stage of our selection process, candidates are preselected through several quality filters:

• Smart Contract Evaluation: A sophisticated scanner is being developed to evaluate the quality of smart contracts associated with each protocol. This scanner checks for uniqueness, potential backdoors, and other security vulnerabilities, ensuring that only projects with secure and innovative technological foundations are selected.

• Interaction and Compliance Checks: Projects that interact with blacklisted addresses are removed from consideration. Additionally, sector-specific characteristics and blockchain compatibility are assessed by our compliance filters, ensuring alignment with a sector-based or chain-specific index (e.g., Arbitrum Top-5 Index).

The core of the multivariate regression results is represented by the second stage. The aim is to collect high-quality and unique data available to most protocols under consideration. Some highly predictive variables may be kept private, and interested parties are encouraged to contact us.

Variables such as the following are included:

• Store of Value:

  • Market cap and fully diluted valuation

  • Total value locked

• Activity:

  • Active wallet addresses and retention

  • Traded volume

  • Smart contract interactions

  • Social media engagement scores

• Technologic Quality:

  • Smart contract unique features

  • Builder activity

Why Elastic Net Regression?

Elastic net regression is leveraged in the second stage of our investment process. This method is particularly suited to financial datasets for several reasons:

• Handling Multicollinearity: Financial markets often exhibit multicollinearity, where multiple variables are closely related (e.g., market cap and trading volume). This issue is mitigated by elastic net regression, which combines the properties of both ridge and lasso regression techniques, enhancing the stability and interpretability of the model.

• Dealing with Heavy-Tailed Distributions: Cryptocurrency market data often shows heavy-tailed distributions, indicating that extreme values are more common than in normal distributions. These are effectively handled by elastic net regression, reducing the risk of model predictions being overly influenced by these outliers.

• Correlated Financial Data: In financial contexts, many predictors might be correlated, which can distort the performance of some models. Elastic net regression's ability to shrink coefficients can help in selecting and correctly weighting the most relevant features, thereby optimizing the investment strategy.

Dynamic Asset Tracking and Index Composition

A wide range of asset signals is tracked by our fund, including traded volume, active addresses, total value locked, and the relationship between market cap and fully diluted market cap. This tracking list is continually refined, with new signals being added as they become relevant. The index composition is then determined by these variables, with predefined minimum and maximum weights per constituent to ensure diversification and manage risk.

By integrating elastic net regression into our investment strategy, our crypto fund not only adapts to the current market conditions but also stays ahead of the curve, potentially outperforming traditional static investment strategies. This dynamic approach, supported by rigorous backtesting and an ongoing commitment to scientific methods, positions our fund as a leader in the innovative application of machine learning in crypto asset management.