Cascade (SN91) Pairs Forecasting with REAX (SN92)-Powered Decisions in Crypto Trading Test

Spotting a breakout is only part of a trading decision. A price can push above its recent high while conditions elsewhere suggest caution, leaving traders to decide whether the opportunity justifies the risk.

Cascade (SN91) explored this challenge by pairing its Ephemeris forecasting API with REAX (SN92)-powered Decision Models. The experiment tested whether anticipating market conditions and adjusting exposure could improve a crypto trading strategy.

Cascade is a subnet developing time-series forecasting models, with Ephemeris providing access to forecasts from historical numerical data.

Decision Models evaluates supplied information and returns structured choices, scores or probabilities.

Their roles in the test were complementary, with Ephemeris forecasting market conditions and Decision Models assessing how much risk to take under them.

The backtest covered BTC, ETH and XRP perpetual futures from January 2024 to June 2026, including actual funding payments and trading costs of five basis points. Its starting strategy was bought when prices closed above the previous 20-day high and shorted below the previous 20-day low. Ephemeris then filtered entries, permitting trades only when it forecast widening Bollinger bands over the next 24 hours.

This indicated an expanding price range in which a breakout might have more room to develop.

That filter increased the Sharpe ratio, a measure of return relative to risk, from 1.08 to 1.49. By comparison, filters checking whether the bands were already widening scored 0.66–0.98 and underperformed the original strategy.

The result suggested that forecasting expansion could be more useful than waiting for it to appear in historical indicators. However, an expected increase in price movement still left the question of how much exposure the portfolio should carry.

To address this, Decision Models assessed broader market conditions each Monday using equities, the VIX, the dollar, US Treasury yields and gold.

It also received recent crypto returns, funding data and seven-day volatility forecasts before assigning a risk-on, neutral or risk-off stance.

Risk-off cut normal gross exposure by half, while daily Ephemeris forecasts continued to filter entries and inform exits. The weekly adjustment improved results at every sizing level tested and beat a simpler risk rule based on the S&P 500 and VIX.

Further tests showed that some uses were more productive than others. Restricting trades to coins expected to trend reduced time in the market from 46% to 20% of days and lowered maximum drawdown, although it also reduced returns.

Forecast-based position sizing performed slightly worse than sizing with historical volatility. Next-day price direction was correct only about half the time, while selecting the following day’s best-performing coin added no value.

The experiment illustrates how two Bittensor services can contribute different parts of an automated trading process, connecting forecasts with explicit risk decisions.

Its results remain preliminary, as analysis across 448 variants did not establish a statistically confident edge after accounting for multiple testing. Fixed-rule paper trading is the next step to assess whether the findings hold beyond the backtest.

For Cascade and REAX, the test provides a concrete application to evaluate further, with recorded forecasts, decisions, and outcomes determining whether the combination delivers lasting value.

A User Used Cascade (SN91) to Predict Their Spotify Wrapped for 2027