At Taoshi, their mission encompasses two main goals. Firstly, they’re committed to democratizing access to advanced quant signals, breaking down barriers that have traditionally restricted them to industry elites. Secondly, they’re determined to demystify the complexities of financial markets, making them accessible to a broad audience—from AI enthusiasts to data scientists and visionary entrepreneurs. They’re dedicated to fostering an intuitive, user-friendly environment that fosters exploration, innovation, and success within the Bittensor ecosystem.

The Challenge and Opportunity of Intraday Forecasting

Intraday price prediction is notoriously challenging, but it’s where significant potential lies. SN8 provides frequent updates, allowing users to make timely decisions on their positions, which is crucial in a market where every second counts.

One of SN8’s most exciting features is its ability to enhance existing models. Quant firms and traders can integrate SN8’s output into their strategies, creating “super models” that are more robust and sophisticated than any individual model.

Focusing on Financial Trading

Competing on SN8 requires miners to have a deep understanding of financial market deep learning models, including feature selection and techniques like high-frequency to low-frequency cascading models and ensemble modeling. Precision is key, and SN8 is where the best compete. Subnet 8 has undergone rebranding to focus on constant financial trading competition among miners, who make trades for various assets and trading pairs.

Miners are assessed based on their trading performance against each other, creating a simulated financial market within bit tensor. The subnet allows individuals to participate in trading without using actual money, making it ideal for showcasing trading skills even with limited funds.

The platform aims to strike a balance between traders making thoughtful, infrequent trades and those providing continuous new insights with higher frequency trades. Different trading behaviors are considered in the evaluation process to encourage diverse participation. Rules are in place to guide traders, including a maximum drawdown percentage and a required number of positions per month. Risk levels are monitored, and deviations from set parameters may result in removal from the system.

Empowering Miners with Taoshi

At Taoshi, they empower miners with the necessary tools for success. They provide access to vetted data sources from top-tier providers across all asset classes. Their Feature Set Creator product transforms raw data into actionable features, helping miners train and refine their deep learning models effectively. With their domain expertise, they continually deliver new versions of open-source deep learning models to foster creativity and competition. These models are available on platforms like Hugging Face, democratizing access to cutting-edge technology.

Previously, the subnet was dedicated to accurately forecasting time series like Bitcoin’s movement, providing valuable predictive insights. To enhance the practical application of predictions, the subnet now focuses on enabling miners to make real trades based on the forecasted information. By directly executing trades based on predictions and evaluating the quality of those trades, the subnet aims to offer a more tangible and beneficial service to users.

Miners are evaluated based on their historical transaction performance, ensuring they exhibit low risk and make valuable trades for future selling. The subnet incentivizes recent trades by prioritizing them, fostering a competitive yet quality-driven atmosphere. With support for multiple assets like Forex, cryptocurrencies, indices, and metals, miners have a wide array of trading options available to explore within the subnet. Miners have the opportunity to specialize in different areas within the marketplace, allowing them to focus on specific signals related to asset classes or trades. By specializing in a subdomain, such as focusing on gold, miners can combine general market movement signals with specific data about that asset for more informed trading decisions.

Theta models build on top-performing miners to create new trading strategies. Signals from top miners provide valuable market insights that can be repackaged for different uses. Theta models aim to optimize the utilization of signals from miners to enhance market analysis.

Guidelines

• They allow miners to submit LONG, SHORT, or FLAT signals across Forex, Crypto, and Equities trading pairs.
• Signals submitted outside of market hours are automatically ignored.
• Miners are only permitted to have one active position per trade pair or symbol at a time.
• Positions are uni-directional—if a position is initiated as LONG, it can’t flip to SHORT mid-way. Trying to do so results in the position being closed, and a new SHORT position must be opened separately.
• Position leverage is limited per trading pair: Minimum order leverage is 0.001, Crypto leverage range: [0.01, 0.5], Forex/Equities leverage range: [0.1, 5]
• Portfolio leverage is capped at 10x across all open positions. Crypto position leverage is scaled 10x when contributing to this cap.
• Miners can take partial profit on a position using opposing signals. Example: To reduce a 0.5x LONG to 0.25x, a 0.25x SHORT signal is submitted.
• To fully close a position, miners simply send a FLAT signal.
• Any miner caught plagiarizing others will be eliminated from the network.
• A spread fee is applied per order and scales with leverage (e.g., 3x leverage = 3x fee).
• A carry fee is applied for open positions: Crypto: 10.95% per year, Equities: 5.25% per year, Forex: 3% per year
• There is a 2.5 TAO registration fee required to join the subnet on mainnet.
• New miners get a 9-day immunity period to submit competitive orders—but eliminated miners cannot benefit from this.
• Miner rewards are based on performance metrics like total return and omega score.

This setup ensures only the most skilled traders and advanced quant systems can succeed in the ecosystem.

Eliminations

Within the Proprietary Trading Network, miners face elimination for engaging in plagiarism. Miners who repeatedly replicate another miner’s trades will be eliminated. Their system assesses the uniqueness of each submitted order, and if an order is identified as a copy (plagiarized), it triggers the miner’s elimination. Following elimination, miners are not immediately deregistered from the network. Instead, they undergo a waiting period, determined by registration timelines and the network’s immunity policy, before official deregistration. Upon official deregistration, the miner forfeits any registration fees paid.

Validators

The Request Network provides the infrastructure for purchasing high-quality predictions curated by the subnet from validators directly. Users can find and purchase signals that match their risk appetite and trading preferences. Validators select trades based on risk metrics provided by miners. Different miners specialize in various market movements, offering a range of risk profiles. Validators match traders with miners based on their risk appetite and trading preferences.

Moss: The Marketplace of Signals

Moss aims to address the need for high-quality data by providing a system where multiple data providers offer their signals, enabling miners to purchase and integrate diverse data sources into their algorithms. This marketplace of signals enhances miners’ ability to access a variety of data streams that can improve their trading models and predictions.

Miners integrate various data sources into their algorithms within the subnet mechanism to enhance trading predictions and make informed decisions based on correlated signals. Moss acts as a source of additional signals that miners can utilize to refine their models and predict market movements more accurately.

The outputs of Theta models can be sourced from various entities such as miners, scientists, or researchers, who may sell their data for predictions like commodity prices. Scientists might share real-time data outputs from satellites through platforms like Moss for others to purchase if they are making predictions related to the data. Accessing high-quality data through Moss could enable other subnets to make informed decisions utilizing the same data architecture.