Five interlocking systems, running every four hours, to find high-conviction Bitcoin entries and exit with discipline before bear markets take hold.
CalyxFi is not a prediction machine. It doesn't try to call tops or bottoms. It does something simpler and more powerful: it identifies when a bull market is structurally in place and signals a long position — then signals an exit the moment that structure breaks down.
This approach has been validated across equities, commodities, and currencies for decades. Bitcoin's halving-driven four-year cycles make it an unusually clean environment for trend-following strategies.
The key insight: missing a -70% bear market is worth more than catching every last percent of the bull. Compounding capital that never suffers a catastrophic drawdown grows faster than capital that does.
Scale varies across panels
The 200-period moving average (roughly 33 days on 4-hour bars) has been Bitcoin's most reliable bull/bear dividing line across every cycle since 2013.
When Bitcoin falls and stays below the MA-200, it signals that the underlying trend has structurally reversed — institutions are distributing, leveraged longs are being liquidated, and the macro tailwind is gone.
CalyxFi doesn't try to pick the exact top. Instead, it uses the MA-200 as an objective, systematic trigger: when price crosses below with a 2% buffer, an exit signal is generated. No emotion. No hesitation.
| Bear Market | BTC Peak→Trough | Signal Status |
|---|---|---|
| 2018–2019 | −84% | In Cash |
| COVID (Mar 2020) | −63% | In Cash |
| 2022 Bear | −77% | In Cash |
Based on historical backtest. Past performance does not guarantee future results.
Avoiding the bear is only half the equation. The other half is being back in position before the bull run accelerates. Miss the first few months of a new cycle and much of the compounding is already gone.
In the 2020–21 bull run, BTC moved from $4k to $12k in the first four months alone. In 2023–24, it doubled within six months of the cycle low. The investors who captured those gains weren't lucky — they had a systematic process for recognising when conditions had structurally shifted back to bull.
The engine monitors the same regime indicators that triggered the exit, watching for the moment the bear structure breaks down. Re-entry signals require multiple independent conditions to align — not just price action, but broader market structure — before a long signal is generated. This layered confirmation is what keeps the system out of bear market bounces and into genuine trend reversals.
| Bull Run | First 3 Months | Full Run |
|---|---|---|
| 2019 Bull | +92% | +191% |
| 2020–21 Bull | +175% | +730% |
| 2023–24 Bull | +60% | +240% |
Missing the first three months of each bull run means giving up roughly half the total gain. Based on historical backtest. Past performance does not guarantee future results.
The engine evaluates new market data on every 4-hour candle close and passes it through a layered decision process before generating a signal.
Every four hours, the pipeline fetches fresh data from multiple sources: Bitcoin OHLCV price data, funding rates, open interest, Fear & Greed Index, cross-asset prices (Gold, S&P 500, VIX), and on-chain metrics.
Raw data is transformed into engineered features across eight layers: technical indicators, volatility measures, microstructure signals, sentiment, cross-asset correlations, on-chain data, and open interest dynamics.
A LightGBM ensemble (direction classifier + volatility regressor) processes the 53-feature set and outputs a confidence-weighted BUY, SELL, or HOLD signal. The model was trained using walk-forward validation across 84 time windows to prevent overfitting.
Even a high-confidence BUY signal is overridden if Bitcoin is in a bear regime (price below the 200-period MA). This single filter is responsible for avoiding every major bear market. A 2% hysteresis buffer on the exit prevents premature exits during normal bull market corrections. An IV/RV spread filter (Deribit DVOL vs realised vol) adds a further gate when options are unusually expensive — a signal of elevated crash risk.
Approved signals pass through a final layer: dynamic position sizing (Kelly-fraction based), ATR-based stop-loss placement, and three-tier circuit breakers that halt trading during extreme market stress. The primary exit mechanism is the MA-200 crossover — the signal holds through the full bull run and closes when the trend breaks. Regime-adaptive take-profit levels (TP1/TP2/TP3) act as a secondary exit layer in ranging markets where the MA-200 exit is less reliable. Execution is via live Kraken market orders.
Markets evolve. So does the engine.
New data sources and engineered features are continuously evaluated — from on-chain flow metrics to derivatives market microstructure. Every candidate feature is tested in isolation before being considered for the model.
No feature or model change ships without clearing the same 84-window walk-forward validation used to build the original model. This guards against overfitting and ensures that improvements are genuine, not curve-fitted to recent data.
The core goal of each improvement cycle is to identify bull and bear regimes earlier and with greater confidence — tightening entry timing at the start of bull runs and sharpening exit signals before drawdowns deepen.
Join the waitlist and be first to access CalyxFi when we open early access.