Subnet 8 :: Proprietary trading Network

This episode unveils Toshi's ambitious endeavor to democratize access to sophisticated financial trading strategies by bridging Subnet 8 (Proprietary Trading Network - PTN) with their user-facing platform, Glitch Financial, offering a transparent, non-custodial gateway to AI-driven alpha.

Episode Introduction and PTN's Genesis

• Host Jake initiated the discussion by reminiscing about early 2023, when Arash from Toshi proactively reached out about building a subnet, months before official support was available. This foresight led to the creation of the Proprietary Trading Network (PTN).
• Jake commended Arash and the Toshi team for their resilience through PTN's "tumultuous time," filled with ups, downs, FUD (Fear, Uncertainty, and Doubt), and FOMO (Fear Of Missing Out), highlighting their continued strength.
• Arash from Toshi outlined their mission: to build the "biggest and best thing to bring the most value to Bittensor." Toshi, with 18 full-time employees and additional contractors, pivoted from an initial prediction subnet to PTN, aiming to create the strongest possible commodity.

Understanding Proprietary Trading

• Paul Liverman, Principal Quant Researcher at Toshi with over 20 years in quant research and trading, explained Proprietary Trading (Prop Trading). This is when a firm uses its own capital to trade financial instruments for profit, assuming full risk and reward.
• Prop firms employ diverse strategies like Statistical Arbitrage (StatArb) – exploiting minute price differences between related assets – quantitative models, and arbitrage, often keeping their methods highly secretive.
• The core value proposition, according to Paul, is delivering "risk-adjusted rewards that are essentially uncorrelated with other assets" and providing a consistent cash flow, traditionally by outsourcing to traders, who are miners in the Bittensor context.

Toshi's Three-Pillar Architecture for PTN

• Thomas, Director of R&D at Toshi, detailed their system's architecture, built on three main pillars:
  1. Subnet 8 (PTN): This is where miners submit their trading signals. These signals are then distilled, ranked, and miners are compensated based on their performance.
  2. Network Model Flow: This layer aggregates and combines the diverse signals from miners in real-time, creating a single, unified trading product.
  3. Glitch Financial: This is Toshi's end-user application, allowing individuals to create accounts and subscribe to the trading strategies curated from the network model's output.
• Thomas emphasized this structure supports the subnet and delivers a tangible product. "This is essentially the overarching flow that we've been creating at Toshi," he stated, setting the stage for a deeper dive into each component.

Deep Dive into Subnet 8: Miner Operations and Evaluation

• Within Subnet 8, miners function like traders, predicting future asset prices (e.g., Bitcoin). They can take long or short positions, with leverage indicating the strength of their conviction.
• PTN currently manages approximately 2,500 predictions per month across various assets.
• Miner performance is evaluated based on risk-adjusted returns, which consider the daily returns of their positions relative to the risk exposure. Key metrics include the Sharpe Ratio, Sortino Ratio, Omega Ratio, and a t-statistic, assessed over a 90-day historical performance period.
• Thomas explained the rationale: "When we're looking to grow the amount of capital that we're eventually going to deploy, we need to maintain a level of balance between risk and return."
• The distributed nature of PTN presents unique aspects: miners can test multiple strategies across different markets (e.g., crypto, FX), an opportunity often siloed in traditional prop firms. While there's a limit on miner numbers, strong performance allows any miner to rise, regardless of background.

The Importance of Risk-Adjusted Rewards and Network Model

• Paul Liverman elaborated on why risk-adjusted rewards are crucial. Unlike raw returns which can be due to luck, risk-adjusted metrics reflect skill, are associated with more capacity, and are "more predictive of future performance."
• This approach separates signal from noise, a fundamental concept in statistical inference and AI. Paul noted, "If you think about the Sharpe ratio... it's essentially like a t-stat from statistics, right? So, it's a signal to noise ratio."
• Toshi uses a weighted average of five different risk measurements because each captures slightly different facets of risk.
• At the portfolio level, the Toshi team focuses on diversification. Thomas highlighted the "rotating door of quality coming in from our miners," necessitating a dynamic network model that can adapt to new talent.
• To achieve this, Toshi has developed a suite of tools, including a backtester for miner portfolios, various optimizers, and is integrating risk and alpha models.

Glitch Financial: Democratizing Access to Trading Strategies

• Arash presented Glitch Financial, Toshi's platform designed to make sophisticated trading strategies accessible. "Really what we're trying to build is like an open hedge fund," he stated, aiming to serve both small and large investors.
• The platform is non-custodial, meaning users connect their existing exchange accounts (e.g., Bybit, Hyperliquid, Binance, DYDX) via API keys that grant trade execution permissions but no fund withdrawal access. Toshi operates this as a SaaS (Software as a Service) product.
• Key features demonstrated included:
  • An overview dashboard consolidating returns across multiple accounts.
  • Strategy selection, where users can subscribe to models derived from PTN.
  • Account aggressiveness settings, allowing users to adjust leverage and risk exposure.
  • Detailed trade history and real-time notifications.
• The flow from PTN miner signal to a Glitch user's trade involves:
  1. Miners submit signals (e.g., 10% portfolio allocation to a Bitcoin long).
  2. Validators receive and process these signals with sub-second latency.
  3. Toshi's R&D infrastructure consumes this data, referencing the network model (a weighted composite of miner strategies).
  4. The system then calculates the trade size for the user's account based on their chosen allocation and aggressiveness multiplier.
• Strategic Implication for Investors: Glitch Financial aims to provide retail investors with access to alpha-generating strategies previously confined to institutional or accredited investors. The non-custodial nature reduces counterparty risk.

Addressing Alpha Decay and Scaling the System

• Jake raised a critical question about alpha decay if strategies become too popular. Arash outlined Toshi's multi-pronged approach:
  • Increasing complexity of strategies developed by participants in their upcoming "Active Validator Program."
  • The inherent difficulty in reverse-engineering strategies due to the "rotating door of miners."
  • A long-term potential to transition to a regulated, custodial model for handling very large capital inflows.
• Arash emphasized the vision: "When we always talk about building on Bittensor, it's about how can we build disruptive systems...we've been called the heretic of Wall Street already by multiple institutional players and I think that's wonderful."

Beta Learnings and Future Collateral System

• Toshi is actively learning from the Glitch Financial beta, focusing on minimizing the price difference between PTN's signaled order price and the actual fill price on a user's exchange account. Latency improvements are key to this.
• Thomas introduced a significant upcoming feature: a collateral system for miners, planned for a preliminary rollout by end of Q2, scaling up in Q3.
  • Miners will post TAO (Bittensor's native token) as collateral on-chain via smart contracts. This collateral will determine the size of their trading portfolio (e.g., $100k in TAO might allow trading a $1M simulated portfolio).
  • This system aims to align miner incentives, encourage focused efforts, and dynamically scale network utility with TAO's value. The collateral is at risk if miners perform poorly.
• Actionable Insight for Researchers/Investors: The collateral system could create a new dynamic where investors can identify and back promising miners by contributing to their collateral pools, effectively creating a marketplace for trading talent on PTN. This also introduces a direct economic stake for miners, potentially enhancing signal quality.
• Paul confirmed this model mirrors practices in traditional prop firms, where track records influence capital-to-collateral ratios.

Network Performance, Strategy Secrecy, and Active Validators

• Regarding the network model's performance, Thomas mentioned backtested annualized returns of 20-35% with a Sharpe ratio between 2 and 3. Live beta is currently focused on infrastructure robustness. Paul stressed that their backtesting includes safeguards against overfitting, such as using out-of-sample data.
• To address concerns about strategy copying in an open system, Toshi is developing an Active Validator Program.
  • Validators will be encouraged to build their own models using Toshi's open-sourced infrastructure and deploy these strategies on Glitch Financial, competing for user delegation.
  • Access to high-quality, low-latency data from top miners will be tiered, potentially based on a validator's stake or delegation size. Thomas also mentioned exploring ZK (Zero-Knowledge proofs) and timelock mechanisms for data integrity and controlled release.
• Strategic Implication: The Active Validator Program could foster a competitive ecosystem for strategy development on top of PTN, driving innovation and potentially increasing the diversity and quality of alpha available through Glitch Financial.

Emission Control and Regulatory Compliance

• Arash addressed the topic of token emission control, confirming PTN will implement a burn mechanism for miner emissions to ensure proportionality between collateral, capacity, and rewards. This burn will reduce as network capacity scales. Toshi will proportionally burn its own ownership emissions.
• On regulatory exposure (SEC, CFTC, etc.), Arash stated Toshi is "very cognizant." Glitch Financial currently restricts access from the US and other specific regions, using IP blocking. They are actively working on regulatory compliance, including hiring for roles like a Director of Investments, to eventually onboard users from these regions, potentially as a Registered Investment Advisor (RIA).

Key Audience Questions and Concluding Thoughts

• The team addressed audience questions on topics like the potential for limit orders on Glitch (on the roadmap), preventing overfitting in models, handling relay attacks (martingale detectors are in place), and the unsuitability of high-frequency trading for the current subnet architecture.
• Jake concluded by reiterating the project's core theme of democratizing access to advanced trading tools and expressed excitement for the real-world validation of PTN's strategies through Glitch Financial user adoption.

Conclusion: A New Frontier for Decentralized Alpha Generation

Toshi's integration of Subnet 8 (PTN) with Glitch Financial represents a significant step towards democratizing sophisticated, AI-driven trading strategies. Crypto AI investors and researchers should monitor the rollout of the collateral system and the Active Validator Program, as these could substantially impact network dynamics, strategy quality, and the overall value proposition of decentralized alpha.