GBP/JPY, nicknamed the Dragon pair, is the highest-ranging major cross pair in the Eaglics model, averaging 140 to 200 pips per session under normal market conditions. It combines the volatility of GBP, driven by BoE policy and UK macro cycles, with the carry-driven dynamics of JPY, which can unwind non-linearly when risk appetite deteriorates.
The Dragon pair's extreme range profile reflects its position as a dual-carry instrument. Sterling carries a positive yield differential over the yen of approximately 375 basis points, the largest spread between any two major-pair currencies in mid-2026. When carry appetite is strong, GBP/JPY trends; when risk-off conditions develop and carry unwinds, the downside moves are exceptionally sharp.
The Eaglics GBP/JPY model applies dual regime classification, BoE policy regime for the GBP side and carry-unwind risk for the JPY side, before generating the high and low band. This two-factor regime approach produces tighter confidence intervals on a pair that most single-architecture models handle poorly.
How the Eaglics 5-Model Ensemble Forecasts GBP/JPY
The structural drivers of GBP/JPY are common knowledge. What is not public is the correct weighting of those drivers against one another on any given morning, and which model architecture is most relevant for the coming session's regime.
The Eaglics ensemble does not apply a fixed formula. All inputs are compiled into an orthogonalized signal library where redundant information is removed before any model processes the data. Regime classification, classifying the prior session's close as low, normal, or elevated volatility, then determines which of the five architectures receives the highest weight in the final high-low band output.
Signal Library Construction
Macro calendar events, cross-asset correlation inputs, realized volatility measures, and inter-session price behavior are compiled into an orthogonalized signal library. Redundant information between inputs is removed before any model touches the data.
Volatility Regime Classification
The Hurst exponent and realized variance metrics classify the prior session's close into one of three states: low, normal, or elevated. This classification is the single most consequential variable in the system, it determines which model architecture is most relevant for the coming session.
Regime-Conditional Model Weighting
LSTM, GRU, Transformer, XGBoost, and Ridge regression outputs are generated independently and then weighted according to each model's historical accuracy within the current regime. Trending models receive higher weight in trending regimes; mean-reversion architectures are upweighted in compressed, low-volatility states.
High and Low Band Generation
The weighted ensemble produces a calibrated high and low band, not a point estimate. The output carries a confidence score reflecting the degree of model agreement, along with the regime tag so subscribers see the market context the system scored the session on.
Pre-Session Delivery
The forecast is delivered to the subscriber dashboard before the London open, when institutional order flow begins positioning for the day's range. Full methodology documentation is available in the Eaglics research framework.