How It Works
Clear and Transparent Cryptocurrency Forecast
Many platforms claim to use artificial intelligence or machine learning to predict markets. But they often do not explain what their models actually do or how accurate their forecasts are. At Teresius AI, we believe in transparency. We explain how our system works and give you the tools to judge our forecasts for yourself.
What Powers Teresius Forecast
Our forecasting system is based on a combination of:
LSTM Neural Networks (Long Short-Term Memory):
This is a deep learning model designed to work with time-series data like price and volume. It is commonly used in finance and is good at recognizing patterns over time.Multifractal Pattern Analysis:
This is a mathematical method that looks at how price movements repeat at different scales. These patterns often appear across time frames in financial markets. Traders often recognize these kinds of repeating shapes by eye. Our system measures and integrates them to help forecast future movement.
By combining these two methods, we create a more flexible and accurate forecasting system.
The Data We Use
We train our models using complete market data:
Open, High, Low, Close, and Volume (OHLCV) for each crypto pair.
Teresius currently provides forecasts for the following six cryptocurrency pairs:
BTCUSDT
ETHUSDT
ADAUSDT
BCHUSDT
BNBUSDT
XRPUSDT
Each forecast can be generated for six different timeframes:
30 minutes
1 hour
2 hours
4 hours
1 day
1 week
We do not offer monthly forecasts. There is not enough long-term historical data in crypto markets to build reliable models for timeframes that long.
What You See in a Forecast
When you choose a crypto pair and a timeframe, Teresius generates a forecast chart. This chart includes:
A forecast channel, which shows the expected range of prices.
A red forecast line, which represents the AI’s predicted close value.
A blue historical line, showing actual close prices from the past.
Candlestick bars, which show real historical volume and price behavior.
A forecast candlestick (in red on the right side), showing the predicted price range for the next period.
The current market price, shown for context.
At the bottom of the chart, we also show:
Forecast error values (called “SIGMA Forecast”) for high, low, and close. These are based on root mean square error.
Market volatility values (called “SIGMA High” or “SIGMA Low”) for comparison. These show how much the price normally varies.
Together, these values help users understand how close our predictions are likely to be, based on both model performance and typical market movement.
Example: 1-Hour and 30-Minute Forecast Comparison
Teresius AI Bitcoin Forecast 1-hour interval
Teresius AI Bitcoin Forecast 30-minute interval
Above are two forecast charts for the BTCUSDT pair, generated on May 15, 2025. The first shows a 1-hour interval, and the second a 30-minute interval. Both forecasts were created within five minutes of each other, which is why the current price displayed on each chart is nearly identical.
You can observe a clear correlation between the forecast channels on both charts. The seven candlesticks on the right side of the 1-hour chart closely reflect the same channel dynamics seen across thirteen candlesticks on the 30-minute chart.
This illustrates how forecast patterns scale across timeframes. This is a result of the system’s use of LSTM modeling and multifractal analysis. You’ll also notice that the Max value on the 1-hour chart is higher than on the 30-minute chart, while the Min value is lower. This reflects the broader expected price range at the longer timeframe, consistent with normal volatility scaling.
Why Forecast History Matters
We believe users should be able to judge our forecasts themselves. That is why we include forecast history directly on the charts.
Each chart shows how previous forecasts compared to real price movement. This makes it possible to:
See how well the forecast matched reality
Build trust in the tool based on past performance
Understand how the system reacts to different types of market movement
We also provide clear numerical values for prediction error and market volatility. This allows both beginners and experienced users to evaluate the usefulness of the forecast.
Why We Use Multifractals
Financial markets often show patterns that repeat at different timeframes. For example, a price dip on a 1-hour chart might look similar in shape to one on a 4-hour chart or even a daily chart. These self-similar patterns are called fractals.
We use multifractal analysis, which means we do not just look for one repeating shape. Instead, we identify and measure many overlapping patterns that appear at different timeframes and levels of intensity. These structures help us improve the forecast by understanding how price movement behaves in complex, layered ways.
This method strengthens the neural network model. It adds a layer of insight that helps the system respond to unusual price activity or market shifts.
Built with Real Constraints in Mind
We are careful not to overpromise. For example:
We only provide forecasts for timeframes where enough historical data exists
We do not claim absolute accuracy
We display both prediction performance and market volatility
We use research-based techniques and continuously improve the system with new data
Our goal is not to “beat the market,” but to provide clear, understandable forecasts that help users make more informed decisions.
Try It Yourself
You will get:
Forecast charts for Bitcoin and other major cryptocurrencies
Access to different timeframes
No registration or login required
You can test Teresius Forecast in seconds. Just open our free Telegram bot:
Who We Are
Teresius AI is developed by Pieoneers Software Inc., a leading Canadian AI and web development company specializing in data-driven solutions.
The project is built in collaboration with academic researchers who specialize in neural networks, time-series forecasting, and complex systems.
We believe AI tools should be open, understandable, and grounded in real data.