This episode unveils Synth Subnet 50's novel approach to financial forecasting, focusing on probabilistic outcomes for AI-driven markets rather than singular price predictions, and explores its implications for the Bittensor ecosystem.
Introducing Synth: Probabilistic Forecasting with Monte Carlo Simulations
- James Ross from Subnet 50 Synth introduces their project, which incentivizes miners to submit Monte Carlo simulations for price forecasts.
- Monte Carlo simulations are computational algorithms that rely on repeated random sampling to obtain numerical results, essentially simulating a multitude of future possibilities rather than a single predicted path.
- Synth's approach differs from traditional price predictions; instead of a single price target (e.g., "Bitcoin will be $112,000 next week"), it provides a probability distribution of potential price paths.
- James explains, "It's saying here's the potential paths of the Bitcoin price... and this is the range of prices that Bitcoin can trade between... and this is the probability distribution that the prices will go down these certain paths."
- This results in a "probability cone" with bands indicating the likelihood of the price falling within certain ranges, focusing on modeling volatility.
- The host notes this "fuzzy" approach seems to align well with the Yuma Consensus, Bittensor's consensus mechanism that rewards diverse and nuanced contributions.
Synth's Current Focus and Future Asset Expansion
- Currently, Synth focuses on BTC (Bitcoin) and ETH (Ethereum).
- James Ross outlines plans to expand to commodities (like gold) and equities (like S&P 500, Nvidia, Google) based on demand.
- He mentions receiving demand from traditional finance institutions for S&P 500 (SPY) and interest in Nvidia and Google.
- Gold is considered a good candidate due to its longer trading timeframes, potentially making it easier to model.
- Crypto assets are advantageous due to their 24/7 market operation.
- There have also been requests for TAO (Bittensor's native token) predictions.
Volatility Trading and the Predictability of TAO-related Tokens
- The host shares an anecdote about a friend who trades volatility using methods akin to Fibonacci analysis, finding the DTA (Dynamic TAO Allocation) subnet token universe highly predictable due to its current market structure.
- This is attributed to the difficulty in trading subnet tokens, lack of leverage, and absence of liquidations that cause sudden price jumps in more mature markets.
- James Ross, however, suggests that TAO-related tokens might be "almost like too volatile right now because it's so early," with daily price changes of 20-50% being common.
- While modelable, he believes it's a bit early, though TAO itself is an interesting asset to model.
- Strategic Implication: Investors should note the nascent and volatile nature of subnet token markets, which may offer unique trading dynamics but also pose significant modeling challenges.
Accessing Synth's Data: Dashboards and Liquidation Probabilities
- Synth's data is publicly accessible via dashboards on their website, synth.co.
- Key dashboards include:
- A volatility dashboard forecasting future volatility and comparing miner forecasts against recent actual volatility.
- A liquidation probability dashboard showing the likelihood of an asset's price changing by 1-5% or more, useful for risk management in perpetual futures trading.
- James Ross explains how this data can warn traders of high liquidation risks, citing an example of a trader with a $100 million leveraged Bitcoin position who got liquidated.
- "We're really trying to do with this data is tell people like, 'hey, there's now like a 80% chance you can get you'll get liquidated in the next 12 hours.'
- The data reveals patterns, such as lower volatility on Sunday nights and higher volatility on Monday mornings (Eastern Time) as US markets open and Asian markets close.
- Actionable Insight: Traders can use Synth's public dashboards to assess short-term volatility and liquidation risks for BTC and ETH, potentially informing their risk management strategies.
Synth's Mining Competition and Data Model
- Synth's competition uses a Continuous Ranked Probability Score (CRPS) system.
- CRPS is a scoring rule that measures the accuracy of a probabilistic forecast, comparing the entire predictive distribution with the actual observed outcome.
- Miners submit tens of thousands of data points (representing 100 potential price paths) every hour, totaling around 160 million data points per day for the subnet.
- Miners are scored on both the calibration (how well probabilities match outcomes) and the concentration of their distribution around actual price movements.
- Public dashboards typically display a "meta model" combining the forecasts of the top 10 miners.
- An API is available for enterprise customers, allowing them to access data from specific miners or custom combinations for backtesting.
- James clarifies that miners submit their predicted price paths, not their underlying models or the raw data (e.g., news sentiment, social media data, historical correlations) used to generate these paths.
- Researcher Note: The sheer volume of data (160 million data points/day) and the competitive CRPS mechanism are designed to refine probabilistic forecasting models continuously.
Comparison with PTN Subnet 8: Different Goals and Audiences
- The host brings up PTN Subnet 8, another prediction engine on Bittensor, and recent community concerns about its emission-to-output ratio.
- James Ross distinguishes Synth by stating they are "trying to solve two different problems."
- PTN Subnet 8 aims to provide specific trading signals for a retail audience via an application (Glitch).
- Synth's goal is to forecast future outcomes to help navigate markets broadly, creating datasets to help manage volatility for any asset.
- Synth's primary end customer is envisioned to be AIs. James states, "Our goal is not to power a financial application. Our goal is to power basically all financial markets... our end customer is going to be potentially like AIs."
- Synth aims to create a versatile data layer for lending protocols, exchanges, and institutional/enterprise clients, rather than focusing on retail trader profits.
- Strategic Implication: Investors should differentiate between subnets based on their target audience and problem scope. Synth's focus on AI-consumable, probabilistic data for institutional use presents a distinct value proposition from retail-focused signal providers.
Synth's Vision: A Data Layer for AIs and Modeling Uncertainty
- James Ross elaborates that Synth's ambition extends beyond financial markets to creating synthetic data for various applications, like traffic flow predictions, offering probabilistic answers.
- Synthetic data is artificially generated data that mimics the statistical properties of real-world data, often used for training AI models, testing systems, or protecting privacy.
- The core idea is to change how humans and future AI agents consume data, enabling them to think probabilistically about future outcomes and manage risk better.
- He emphasizes that modeling asset prices is just the "beginning of us like trying to model uncertainty across like all markets."
- The host aptly rephrases Synth's identity: "You're not price prediction subnet, you're probability cloud technology subnet... It's not about trading. It's not about prediction. It's about probability cloud management."
Accuracy Benchmarks and Enterprise Focus
- Synth benchmarks its models against the Geometric Brownian Motion model, a standard model for asset prices, particularly in options trading.
- Geometric Brownian Motion (GBM) is a continuous-time stochastic process often used in mathematical finance to model stock prices.
- James reports that Synth is currently outperforming this benchmark by roughly 25%, an improvement from 20% a month prior.
- This outperformance is a key metric for attracting enterprise customers, including hedge funds who might use the data or even benchmark their own models against Synth's miners.
- James notes, "That's very much our focus like improving on these benchmarks month over month and the way that like Bittensor's incentive mechanism works, it's like really helping us to do that."
- An interesting use case mentioned is that Synth's data could potentially be used to mine other price prediction subnets, thereby improving the broader ecosystem.
James Ross's Background: From Fintech to Mode Network and Synth
- James Ross studied economics, worked in finance in London, and built FinTech companies, one focused on data analysis.
- He entered crypto around 2016-2017, working with Ethereum community teams.
- He co-founded Mode Network, an L2 (Layer 2) network on Ethereum.
- L2 Network (Layer 2 Network) refers to a secondary framework or protocol built on top of an existing blockchain (Layer 1) to improve scalability and efficiency.
- Mode's focus on finance, AI, and agents (AI entities transacting on-chain) led to the realization that a crucial data layer for these agents was missing. This insight spurred the development of Synth.
- Alaa, a Bittensor founder, provided crucial feedback on Synth's concept and scoring mechanism during a Mode demo day.
- Synth launched in January. While sharing some team members, Mode and Synth are separate products, with Mode utilizing Synth's data in its L2 exchange and AI terminal.
- Synth's primary product is its API, targeting enterprise clients.
Discussion on Bittensor's DTA: Is it Working?
- The host raises the question of whether DTA (Dynamic TAO Allocation), Bittensor's mechanism for directing TAO emissions to subnets, is functioning as intended.
- Concerns exist about subnets like Taoshi (a Bitcoin mining subnet) receiving high APY despite not being "traditional AI."
- James Ross, drawing from Constant (likely referring to a prominent Bittensor figure, possibly Const, a core developer), emphasizes that DTA is a "long-term game."
- He believes value will accrue to subnets that consistently innovate and deliver market value.
- He anticipates newer, innovative teams will rise to top positions over the next three months, displacing some early leaders who may have benefited from initial ecosystem support.
- Investor Insight: The DTA landscape is expected to be dynamic. Investors should look beyond current emission leaders to identify subnets with strong fundamentals and continuous innovation, as these are likely to capture value long-term.
Root APY: Too High?
- The discussion shifts to the Root APY, the staking reward for TAO staked directly to the root network, currently around 25% (at the time of recording, though James mentions 48% which has since decayed). Only about 7% of TAO participates in DTA for subnet tokens.
- James Ross opines that the decay of the Root APY "should be faster or slightly faster."
- He suggests, "I think it would be better if it moved to 20% over the next 3 months and then like finds like a lower level."
- He believes reducing Root APY more quickly would stimulate more activity and capital flow into subnets.
- The host advocates for a more drastic cut, suggesting Root APY should be around 3% to incentivize capital movement out of passive root staking and into active subnet participation.
- Strategic Consideration: A high Root APY may disincentivize investment in subnets. Crypto AI researchers and investors should monitor discussions and potential changes to Root APY, as this could significantly impact capital allocation within the Bittensor ecosystem.
The 128 Subnet Freeze and Competition
- The host asks about the idea of a freeze at 128 subnets, with subsequent deregistration and liquidation of underperforming ones.
- James Ross supports this, recalling the "mad scramble" to secure validator weights before DTA to avoid deregistration.
- He states, "I feel that like everyone should go through that stress. I feel like it's a right of passage as a subnet to compete."
- He argues that just as miners face constant competition, subnets should too, fostering a healthier, more competitive ecosystem.
- He believes 128 subnet slots are reasonable and that underperforming teams should be deregistered to make way for innovation.
- Ecosystem Health: Limiting subnet slots and enforcing performance-based deregistration could drive higher quality and innovation within Bittensor, a factor for long-term investors to consider.
Reflective and Strategic Conclusion
- Synth's probabilistic forecasting for AI-consumable data marks a shift from simple price prediction, aiming to become a foundational layer for risk management in diverse markets.
- Investors and researchers should monitor Synth's enterprise adoption and its influence on how AI agents interact with financial data.
