This episode of Novelty Search offers a deep dive into Tashi's ambitious project to bridge decentralized AI-driven trading signals from the BitTensor network with real-world financial markets through their Glitch Financial platform, revealing a new frontier for Crypto AI investors.
The Journey of PTN and Tashi's Vision
- The discussion kicks off with Jake, the host, and Arash, from Tashi, reminiscing about the early days of BitTensor subnets in 2023, highlighting the perseverance of the Prophet (PTN) subnet team.
- Arash, whose company Tawi employs 18 full-time staff plus contractors, emphasizes their commitment: "We want to build this to be the biggest and best thing to bring the most value to BitTensor."
- The presentation aims to cover what Tashi is building with PTN (Subnet 8), the incentive mechanisms, the output's value, scaling strategies, and their overall vision.
Understanding Proprietary Trading
- Paul, from Tashi, provides a concise introduction to proprietary trading (prop trading). This is when a firm trades financial instruments with its own money, as opposed to customers' money, to make a profit for itself.
- He explains that prop trading firms aim to generate "risk-adjusted rewards that are essentially uncorrelated with other assets," employing sophisticated strategies and risk management. In Tashi's model, miners on the subnet act like traders for the firm.
Tashi's Three Pillars: Subnet, Network Model, and Glitch Financial
- Thomas, Director of R&D at Tashi, outlines their three-pillar architecture:
- Subnet 8 (PTN): Where miners submit predictive signals.
- Network Model Flow: Aggregates and refines miner signals in real-time.
- Glitch Financial: The product delivering these curated strategies to end-users.
- Subnet 8: The Foundation of Prediction
- Miners on PTN predict future prices (e.g., Bitcoin), submitting long or short positions with selectable leverage. Leverage here indicates the strength of belief and position size.
- PTN currently manages approximately 2,500 predictions per month across various assets.
- Daily returns from these positions are used to evaluate miner quality over time.
- The Importance of Risk-Adjusted Returns
- Miner performance is evaluated using risk-adjusted return metrics over a 90-day period. These include the Sharpe Ratio (measures return per unit of risk relative to a risk-free asset), Sortino Ratio (similar to Sharpe, but only considers downside volatility), Omega Ratio (probability-weighted ratio of gains versus losses), and a t-statistic (measures the statistical significance of returns).
- Paul emphasizes, "risk-adjusted measurements reflect skill and often are associated with more capacity and perhaps but most importantly are more predictive of future performance."
- Unique challenges in a distributed system include managing diverse miner strategies and the open nature of participation, which Tashi views as a double-edged sword offering both opportunity and complexity.
The Network Model: Aggregating Intelligence
- Paul and Thomas discuss the network model, designed to optimize portfolio-level risk-adjusted returns by considering the uniqueness and correlation of miner signals.
- Tashi employs a custom backtesting engine, optimizers, risk models, and alpha models. Alpha (in finance) refers to the excess return of an investment relative to the return of a benchmark index.
- A key challenge is the "rotating door of quality" as new miners emerge, requiring a dynamic network model.
- A demo showcased a backtested crypto-specific model, illustrating how it composes outputs from various miners in real-time, with plans to open-source parts of this for community model building.
Glitch Financial: Democratizing Access to Sophisticated Trading
- Arash introduces Glitch Financial, Tashi's platform aiming to be an "open hedge fund." It provides users access to strategies derived from the PTN subnet and Tashi's network model.
- Platform Overview and Features:
- The platform is currently in beta, onboarding users gradually.
- It offers account overviews, strategy management, trade history, real-time notifications, and a help desk.
- Users can connect their existing accounts from exchanges like Woo X, Bybit, Hyperliquid, DYDX, and Binance.
- Glitch Financial operates on a non-custodial basis, meaning Tashi never holds user funds. Users grant permission for trade execution via API (Application Programming Interface) keys with restricted access. It's a SaaS (Software as a Service) product.
- Users can adjust "account aggressiveness" (leverage) and "order fill aggressiveness."
- The Mechanics: From Miner Signal to User Account
- Miners send signals to PTN validators.
- Tashi's infrastructure processes these signals with sub-second latency, feeding them into their R&D pipeline (which uses Kafka, a distributed event streaming platform).
- The network model, a weighted aggregation of miner signals, determines trade parameters.
- Glitch Financial bots then execute trades on users' connected exchange accounts, scaled to their chosen account size and aggressiveness settings.
- Vision and Addressing Alpha Decay
- Arash states the mission: "to build something that actually works for people that they can rely on," offering retail investors access previously limited to accredited investors.
- To combat potential alpha decay (the erosion of a trading strategy's profitability as it becomes widely known), Tashi relies on the increasing complexity of strategies, the constant influx of new miner talent, the upcoming Active Validator Program, and a long-term plan to potentially offer custodial services under appropriate regulatory frameworks.
Performance, Latency, and Continuous Improvement
- Arash highlights that a key focus during the Glitch Financial beta is improving execution quality.
- They are meticulously tracking the price difference between theoretical fills on PTN and actual fills on Glitch Financial, aiming for this "slippage" to be less than 0.1% for crypto markets.
- Latency reduction across the pipeline (PTN data streaming -> R&D infrastructure -> Glitch Financial bots) is a primary goal.
Scaling with Collateral: Aligning Incentives and Capital
- Thomas explains the plan to introduce a collateral system to scale PTN's utility. Currently, miners trade with simulated $100,000 accounts.
- The Collateral System Explained:
- Miners will be required to post Tao (referred to as 'theta' by the speakers), BitTensor's native token, as collateral via smart contracts. This collateral will determine their effective trading portfolio size.
- For example, with a 10% drawdown limit, a miner would post 10% of their desired capital base as collateral.
- This system aims to align miner incentives with performance, encourage focused efforts, and allow the network's utility to scale with the value of Tao.
- Rollout is planned for late Q2/Q3.
- Fostering a Competitive Miner Ecosystem:
- Investors will be able to contribute to miners' collateral pools, speculatively backing promising talent and increasing their trading capacity.
- Arash envisions "per market subnets" and tiered leagues (e.g., Premier, A, B, C) for miners, based on collateral and performance, creating a clear path for progression.
- Paul notes that requiring traders to have "skin in the game" via collateral is common in traditional prop trading.
Network Model Performance and Real-World Application
- Regarding the network model's performance, Thomas mentions backtested results showing 20-35% annualized returns with a Sharpe ratio between 2 and 3.
- Paul elaborates that their backtesting incorporates robust measures to avoid overfitting, such as using optimizers on prior returns and out-of-sample testing. The model uses factors like momentum, mean reversion, and Bitcoin beta.
- Arash confirms that the Glitch Financial demo account shown uses real capital, but reiterates that the current beta phase prioritizes ironing out bugs and latency issues over immediate profit and loss (P&L) demonstration.
The Active Validator Program: Enhancing Utility and Competition
- To address concerns about strategy copying and maintain competitive edge, Tashi is introducing an "Active Validator Program."
- Validators will compete to provide the most utility (e.g., by building superior models on Tashi's open-sourced infrastructure).
- Higher-performing validators will gain priority access to the highest quality and lowest latency signals from miners.
- These validators can then offer their derived strategies on Glitch Financial, creating a competitive marketplace.
- Access to miner data and capacity will be tied to a validator's delegated stake, incentivizing validators to attract Tao holders. Tashi's subnet ownership emissions will help fund this program. Quant team SN88 is an early participant.
Navigating Emission Burns and Regulatory Landscapes
- Emission Management for Sustainability:
- Arash discusses the concept of burning a portion of miner emissions, linking it to the balance between collateral, network capacity, and rewards.
- The idea is that as the network's capacity (driven by collateral) scales, the necessity for such burns on miner emissions will decrease, with an expectation for this by Q4.
- Tashi has committed to proportionally burning its own subnet ownership emissions, a decision made following a community town hall.
- Regulatory Compliance and Expansion:
- Arash confirms that Glitch Financial currently restricts access from the US and other specific regions, employing IP blocking.
- They are actively working to bring on legal expertise to navigate regulations from bodies like the SEC (Securities and Exchange Commission) and CFTC (Commodity Futures Trading Commission), and to become a RIA (Registered Investment Advisor), with the goal of legitimately onboarding US users and expanding to other restricted regions.
Audience Q&A: Key Concerns and Future Directions
- Value Accrual, Limit Orders, and Overfitting:
- Value for Tashi's ecosystem (and potentially an associated token) will primarily come from Glitch Financial's operations.
- Limit orders are on the roadmap and acknowledged as important for institutional-grade strategies.
- Concerns about overfitting in backtests were addressed by Paul, emphasizing out-of-sample testing and disciplined model development.
- Data Integrity, Relay Attacks, and HFT:
- Thomas mentioned exploring ZK (Zero-Knowledge) proofs for data integrity. Jake suggested time-locking, which Thomas acknowledged.
- Relay attacks (where one entity copies another's signals) are an ongoing challenge. Tashi has implemented a martingale strategy detector (a high-risk strategy of doubling down on losses) and is working on orthogonality measures to disincentivize highly correlated, unoriginal strategies.
- High-Frequency Trading (HFT) is not supported or targeted by PTN due to inherent blockchain latencies.
- Strategy Examples and Data Access:
- Arash mentioned one of their models showing a roughly 2:1 return-to-drawdown ratio in live (low leverage) beta testing, while cautioning that bugs are still being addressed.
- Regarding data leakage, the Active Validator Program is designed to make top-tier data a competitive asset, less likely to be openly shared by leading validators.
Reflective and Strategic Conclusion
This discussion reveals Tashi's comprehensive strategy to translate AI-driven predictions from BitTensor's Subnet 8 into a tangible financial product via Glitch Financial. Crypto AI investors and researchers should monitor Glitch Financial's beta progress, the rollout of the collateral system, and the Active Validator Program, as these will be key indicators of the platform's viability and its potential to disrupt access to sophisticated trading strategies.
