Week of 6 Oct 2026
A week defined by two competing forces: oil revenue anxiety dragging the naira, and a CBN that held its rate and its nerve. Here is what moved our signals and what we are watching into next week.
What happened this week
Brent crude fell 3.2% on demand revision forecasts from the IEA. Nigeria's FX inflows are oil-denominated — less oil revenue means less dollar supply at the CBN window. USDNGN drifted from ~1,580 to ~1,611 through the week. Our macro regime signal flagged elevated uncertainty on Tuesday; we cut confidence thresholds on NGN signals for 48h.
The CBN held its benchmark rate at 26.75%, in line with consensus. No surprises. Our policy signal had priced a hold at 87% confidence from Monday — the decision itself produced minimal USDNGN movement, which is the correct outcome when the model reads the setup correctly.
Despite the FX headwind, industrials on the NGX showed quiet strength. DANGCEM.LG volume ratio hit 1.8× on Thursday — one of the cleanest liquidity gate passes we have seen on a Nigerian equity this month. Our sector flow feature caught rotation from financials into cement and building materials early in the week.
Crypto was the clearest winner. BTC held $62k support and pushed toward $65.4k by Friday. Our momentum signal on BTC went decisive-long on Wednesday morning with 7/9 gates passed; the 1h outcome confirmed. ETH followed with a lag, which is a pattern our inter-asset correlation feature is beginning to learn.
"The losing calls this week were concentrated in pegged-FX pairs on Tuesday, when the oil price drop was faster than our regime signal could re-weight. That's a known lag. We're shortening the re-weight window."
Where we were wrong
5 of the 13 losing calls this week were in the USDNGN and pegged-FX cluster on Tuesday 7th. The speed of the Brent move (down 2.1% in under 90 minutes) triggered a regime shift faster than our weighting update cycle. We are shortening the macro re-weight window from 4h to 90 minutes, which should catch moves like this in future. We publish this because knowing where the model lags is more useful than reading a win-rate number in isolation.