AlphaNet Beta R2 Pre-Release Notes

As the very much awaited second version of AlphaNet (Beta R2) is set to launch imminently, we would like to give a sneak peek of what’s in store. As the first Public Beta Release (R1) was merely a teaser of a few basic tools that will remain free, Beta R2 will be a significant upgrade and serves to showcase some of the more advanced capabilities of what AlphaNet can provide.

In Beta R2 we will introduce and release the first AlphaNet Hybrid Staking module – it will turn the Hybrid Staking proof-of-concept into an actual practice that we will see taking effect on Phoenix’s token economics and demand for PHB. For those that are not familiar, Hybrid Staking is an innovative approach to staking that at its core is staking in exchange for value or capabilities, instead of only token rewards. Hybrid Staking is a Phoenix ecosystem-wide initiative that extends beyond AlphaNet into Computation Layer and other modules. Read more on Hybrid Staking here: https://token-econ-update.phoenix.global

AlphaNet is positioned as an AI platform for the most advanced retail traders – but with the R2 release it will be easy to see how it is an alpha and profit generation machine in various market conditions, a necessity in the crypto market. At AlphaNet, we focus on quality over quantity – we provide users curated products with cutting edge AI combined with market expertise to provide significant value creation. In these pre-release notes, we will introduce some of the new offerings we will have in Beta R2.

ViperAI

ViperAI is an advanced AI model that delivers superior risk-adjusted return by exploiting market inefficiencies in micro-trends from 20 minutes to multi-hourly timeframes. ViperAI is a long-short market neutral strategy (maintains stable performance in various market conditions), and maintains a highly asymmetric per-trade RR (risk-reward). This system is able to capture both directional trade opportunities as well as strong momentum reversal signals, allowing the trader to capitalize upon volatility in both directions.

ViperAI utilizes over 140 features (data points/factors) from granular high frequency futures data including those derived from price, trade data (transactions), volume, maker/taker data, orderbook and more. The model has been trained with market data from the past 18 months and has been backtested for 12 months, with an additional 2 months of trading in live. Technologies used include multiple layers of LSTM (deep neural networks) with XGBoost, with Deep Reinforcement Learning (AI tech used for AlphaGo) for signal parameter tuning. This strategy is built using over a span of 6 months and over 7200 man-hours of research & development, and serves as the first flagship trading strategy offering for AlphaNet users.

Let’s take a look at the performance of ViperAI for BTC futures, which is arguably the most competitive, most difficult to game, most market efficient asset in the market:

These numbers include the cost for Binance Futures trading fees, averaging maker/taker fees, and the strategy does not use any leverage. The performance for the single asset class of BTC already surpasses the risk adjusted return of vast majority of crypto funds as well as most crypto trading firms. ViperAI performance for different altcoin assets are expected to have more edge than BTC and surpass its performance metrics.

Here is a breakdown for the monthly return on BTC futures:

Click below link to see ViperAI in action for a sample month (June 2023):

https://alphanet.phoenix.global/files/ViperAI-06-2023.html

In addition to the default ViperAI timeframe profile that results in average ~100 trades per month, there are longer trading timeframe versions of ViperAI that have a different signal generation mechanisms targeted towards traders of a different trading profile. These will be available gradually after Beta R2 launch. Here’s how they look like:

As you see, these two variations of ViperAI deliver similar results for BTC, but with different signals and slightly different drawdown profile, with L2 only having 12 signals over a 7 month period and M1 having 41 signals. Note that with these results average exposure of 60% was used with no compounding – these results do not apply leverage nor max capital exposure.

We are currently in the process of integrating Telegram notifications so that users can easily get notified on low-frequency signals. ViperAI will be a continuous project that will upgrade technology and add new trading pairs frequently.

Access to Viper AI: Hybrid Staking

ViperAI will not have any direct subscription cost and Hybrid Staking via AlphaNet will be required. Users who wish to derive alpha from profitable AI capabilities on AlphaNet such as ViperAI must participate in the AlphaNet token economics by acquiring and staking PHB. Our AlphaNet token economics litepaper will be released shortly, detailing why Hybrid Staking will create a virtuous value cycle and preserve alpha creation.

WaveML

WaveML is an Insights offering on AlphaNet that enables users to detect inefficiencies in short-term trends and market dynamics that enable trade opportunities. WaveML is built on the core principle that stands as the antithesis of the Efficient Markets Hypothesis (the notion that markets are efficient and that little to no edge can be exploited). WaveML helps traders identify structural pockets, or “waves”, that can be used to structure trade opportunities.

The fundamental building block of WaveML is a structure called a Wave – here we are not referring to Elliot Waves but a new AI-based pattern structure developed by researchers at Tensor Investment. The premise of the Wave depends heavily on patterns in volatility, volume, and price action, and through AI models learns the relationships through these 3 main factors. The start of a wave marks the signal of exiting rangebound-like regime and entering into volatile directional regime. However, the key details lie in how the transition into a Wave occurs:

Direct transition (Phase 0 => 4): This typically represents a market-efficient transition that likely has no alpha. If a trader executes a directional trade upon this type of transition, whiplash or instant reversal is highly probable.
Step transition (Phase 1=>2=>3): This type of transition signifies a likely less efficient market state, hence a higher probability for “alpha” and more ripe for direction trade execution opportunity.
Continuation (Phase 3=>4, 4=>5): This is a continuation detected within a wave – presents short-timeframe continuation trades.

Another important use case of WaveML is for calculating volume profiles and VWAP (volume weighted price average) lines. Traditional calculation methods utilize a fixed time window (ie. 30, 60, 120 minutes), but much of this is noise and the fixed-window approach does not account what weight of different data points should play in the calculation. Via WaveML, users can calculate a wave-based volume profile and VWAP values that give much more edge and statistical significance based on AI.

Implementation of WaveML included 3 components:

A volatility prediction module using deep learning (LSTM), utilizing over 100+ data points (features)
8 directional prediction modules utilizing a combination of deep learning and XGBoost, utilizing over 180+ data points.
A Hidden Markov Model (HMM) for detecting transition states between volatility and directional model outputs.

Access to WaveML: Hybrid Staking

Access to WaveML will not have a subscription cost, and Hybrid Staking via AlphaNet is required.

Note that Hybrid Staking tiers T1, T2, T3, etc are AlphaNet universal tiers – meaning that once you stake at a particular tier, you can access all products and offerings with that tier requirement.

Last updated 1 year ago