Understanding DeFiStar's Experimental Prediction System | Learn How to Interpret Signals, Validate Strategies, and Understand Verification Timelines
DeFiStar's Prediction Engine V1 is an experimental research framework designed to help you validate DeFi and Forex investment strategies by tracking historical patterns and making probabilistic forecasts. Think of it as a laboratory notebook that shows you what's happening on-chain/in-markets and tests hypotheses about asset behaviour.
Use this system to answer questions like: "Do yields actually stay stable for established protocols?" or "Does high utilisation really lead to rate spikes?" or "Can neural networks identify currency reversal patterns better than indicators?" These are research questions, not trading signals.
Each prediction forecasts what will happen in the NEXT X days from the moment it's generated. This is NOT historical analysisâthese are live predictions about the future. The "verification timeline" is the period we're predicting, not when we check old data.
When the system makes a prediction on 29 December 2025 at 11:00 UTC, here's what happens:
Today (29 Dec 11:00): System predicts "USDC will heat up over the next 3-5 days"
What it's forecasting: APY will increase between now and 1-3 January
Verification date: 3 January 2026 (5 days later)
Verification process: System checks if APY actually increased during 29 Dec â 3 Jan period
Outcome: Correct â or Incorrect â
| Engine | Prediction Window | What Period Is Being Forecast | Why This Timeframe |
|---|---|---|---|
| APY Momentum | Next 3-5 Days | Short-term APY direction (will it rise or fall in the next 3-5 days?) | Momentum signals fade quickly. Longer windows introduce too much noise. |
| Reliability Index | Next 7-14 Days | Will consistency hold over the next 1-2 weeks? | Reliability is a medium-term metric. Need sufficient time to see if stable yields persist. |
| TVL Momentum | Next 3-5 Days | Capital flow direction over the next few days (inflow or outflow?) | Money moves fast in DeFi. TVL shifts are observable quickly. |
| Utilisation Forecast | Next 7 Days | Will current utilisation level trigger rate changes in the next week? | Rate changes from utilisation spikes take days to fully manifest. |
| Signal Confluence | Next 2 Days | Fast-track directional forecast for the next 48 hours | Multi-signal strategies need quick validation. |
| GARCH-LSTM | Next 24 Hours | Short-term currency rate and volatility forecast | Forex markets require high-frequency updates due to rapid price discovery. |
Short-term strategies (24h - 5 days): Use GARCH-LSTM, APY Momentum, TVL Momentum, Signal Confluence. These engines predict rapid changes.
Medium-term strategies (1-2 weeks): Use Utilisation Forecast and Reliability Index. These capture slower-moving trends.
Long-term strategies (>2 weeks): This system is NOT designed for long-term forecasts.
The Predictive Dashboard is your central command centre. It shows real-time accuracy metrics across all prediction engines and displays only pools with sufficient historical data for reliable analysis.
| Metric | What It Means | How to Interpret |
|---|---|---|
| System Accuracy | Percentage of verified predictions that were correct | Higher is better. 50% = coinflip. 70%+ = strong signal. |
| Verified Predictions | Total number of predictions that have passed their verification date | More verified predictions = more reliable accuracy metric. |
| MAE (Mean Absolute Error) | Average magnitude of prediction errors | Lower is better. <0.5% = excellent. >2% = noisy signal. |
| Confidence Score | MAE-adjusted probability (0-100%) | Higher = system has been more accurate for this asset historically. |
Short-term APY momentum (heating up or cooling down) based on Exponential Moving Average (EMA) analysis.
The probability that a pool will maintain its current yield consistency over the medium term.
Capital flow direction (Liquidity Migration) based on net TVL changes.
Near-term APY movements derived from changes in protocol borrow-utilisation.
High-conviction directional forecasts when multiple engines (APY + TVL + Utilisation) align.
Advanced statistical and neural network forecasting for Forex markets, predicting currency exchange rates and volatility clusters.
Methodology: This engine combines GARCH (Generalized Autoregressive Conditional Heteroskedasticity) models to forecast volatility with LSTM (Long Short-Term Memory) neural networks to predict price direction.
| Signal | Logic | Action Forecast |
|---|---|---|
| đ BUY | Neural consensus indicates upward momentum with stable volatility. | Target rate increase within 24 hours. |
| đ SELL | LSTM detects reversal patterns or downward trend continuation. | Target rate decrease within 24 hours. |
| â¸ď¸ WAIT | High GARCH-modelled volatility or neutral neural signals. | No clear directional edge identified. |
The engine requires at least 200 historical data points to generate an LSTM forecast. It uses MinMaxScaler to normalise currency rates and employs Dropout layers to prevent overfitting to noise.
A signal is not a command. It is a probabilistic data point. Here is how to use them effectively for strategy validation:
Transparency is our core principle. We don't hide failuresâyou can see exactly which predictions succeeded and failed.
Currently, DeFi accuracy ranges from 25% to 66%. The GARCH-LSTM engine targets 85-92%. These numbers change daily as more predictions are verified.
DeFi is volatile. Black swan events, governance changes, and massive capital moves can invalidate any model instantly. Use these for strategy validation, not certainties.
The prediction engine and all associated analytics are experimental. DeFi protocols can change parameters at any time, invalidating all forecasts. Market conditions can shift rapidly and unpredictably. Use this tool for educational research and strategy validation only. Do not make financial decisions based solely on these signals. You are responsible for your own due diligence and risk management. See our Terms & Conditions for full legal disclaimers.