Currency strength,
as institutions see it
Eight major currencies. Four timeframes. BIS 2025 volume-weighted. XGBoost conviction scoring. Updated every H1 bar from a live MT5 feed — the methodology made real.
From concept to running system
How do institutions measure currency strength? Not equally — volume-weighted across timeframes. BIS 2025 turnover data provides the weighting. H1, H4, D1 and W1 each carry a different analytical role, anchored by a 48-bar moving average that defines the prevailing trend context. The result is WA-RSI: a single number per currency that reflects real market participation and trend direction together.
Read Article #2 →WA-RSI divergence identifies pairs. XGBoost trained on 33 walk-forward splits predicts MA48 trend persistence — will a currency's strength continue over the next 24 hours? Signal accuracy of 67% against a 33% random baseline. The model acts as a trend confirmer, not a contrarian predictor. Only the highest-conviction calls are shown — not every signal is worth acting on.
Read Article #3 →CoreFX Engine EA computes WA-RSI and its 24-bar and 48-bar moving averages from your MT5 broker feed every H1 bar. The Python ML pipeline scores trend persistence conviction using a 96-feature XGBoost ensemble. CoreFX Web and the Windows desktop app display both — WA-RSI strength bars with MA trend context and ML conviction trades — updated every H1 bar close.
You are hereHow to read the dashboard
The WA-RSI scale runs from 0 to 100. 50 is neutral. Above 52 is strength. Below 48 is weakness. The ML model predicts whether that strength or weakness will persist over the next 24 hours based on each currency's trend relative to its 48-bar moving average.
Each bar is a currency's Weighted Average RSI across H1, H4, D1 and W1 timeframes, scaled by its BIS 2025 global FX turnover share. Green bars above 52 indicate strength. Red bars below 48 indicate weakness. Bars between 48 and 52 are neutral — the ML may already have a view even when the bar is in this zone. The arrow shows direction vs the previous H1 bar close.
Each badge shows the ML model's directional conviction. BUY % — ML predicts the currency will stay above its 48-bar MA. SELL % — ML predicts it will stay below. The badge shows whenever the model fires, regardless of where WA-RSI currently sits. No ML means the model produced no signal above threshold this bar — the dim BUY/SELL comes from WA-RSI position only.
Two distinct conviction numbers are shown. The currency conviction (badge %) is the XGBoost model's calibrated probability that the currency's trend relative to its MA48 will persist over the next 24 hours. The pair conviction on the trade card is the minimum of both legs — conservative by design, since the trade is only as reliable as its weakest currency signal. A 67% pair conviction means the model was correct 67% of the time on similar setups historically.
Conviction trades appear when both legs have ML signals above their per-currency threshold. ALIGNED — WA-RSI and ML agree on both legs: the trend is confirmed. MIXED — the ML has fired but one or both WA-RSI readings have not yet crossed their threshold (48 or 52). MIXED is not a failed signal — it means the ML is leading the price action. The trade card shows exactly which leg needs to confirm and what level to watch. When that level breaks, the signal becomes aligned.
What drives the numbers
The calculation behind every bar, every badge, and every signal.
Four timeframes, each carrying a different analytical role. H4 carries the most weight — it captures the session-level trend that drives intraday setups. W1 acts as a structural veto, suppressing contra-trend signals when the weekly bias is clear.
(H1 × 0.30) + (H4 × 0.35) + (D1 × 0.20) + (W1 × 0.15)
scaled by BIS 2025 volume weight, normalised across 8 currencies
Not all currencies trade with equal weight. Raw BIS 2025 turnover shares are normalised across the 8 majors so they sum to 100% — this is the actual weighting applied in the WA-RSI calculation.
Raw BIS shares (USD 89.2%, EUR 28.9%...) sum to over 200% because every FX transaction involves two currencies. Normalising to 100% gives the proportional weight each currency carries in the WA-RSI calculation.
A separate XGBoost model is trained for each of the 8 currencies. Each model predicts MA48 trend persistence — will this currency's WA-RSI remain on the same side of its 48-bar moving average 24 hours from now? This makes the model a trend confirmer rather than a mean-reversion predictor. Training uses 33 walk-forward splits across approximately three years of H1 data, with isotonic regression calibration so conviction percentages are true probabilities rather than raw model scores.
The model answers: "Is this currency trending above or below its 48-bar (2-day) moving average, and will that relationship hold for the next 24 hours?" BUY = above MA48 and expected to stay above. SELL = below MA48 and expected to stay below. HOLD = near the MA or about to cross. This is why MIXED trade cards show specific levels to watch — the ML has fired, but the WA-RSI crossing its threshold (48 or 52) is the confirmation signal.