Forecast

InHouse gold forecast: GARCH cone and expected path

Gold price cone from GARCH simulations (bands at 5%, 33%, 66% and 95%) with the expected path in blue, and the odds of each zone below. Pick a report date to see a past forecast and the price that followed.

Method. Four GJR-GARCH(1,1) models with Student-t errors (lookbacks of 1, 2, 5 and 10 years) simulate 1,000 paths to 21, 63, 126 and 252 business days; the quantiles at those points are joined with a pchip curve. Bands: 5%-33% (red), 33%-66% (gray), 66%-95% (green). Blue line: expected path = band midpoints (the median for Neutral) weighted by the probability of each zone. Bottom: zone probabilities from the ordered logit at 30, 90, 180 and 365 days, interpolated daily from an even split on the report date. The chart ends 252 business days out, so its end labels are interpolated and can differ by about 1 point from the 365-day odds. The simulation has no seed: each cone is simulated once and kept, so a fresh notebook run gives slightly different numbers.

Zone probabilities by horizon

Report date 12 Oct 2026. Each horizon scores its own window after the report date: days 1-29, 30-89, 90-179 and 180-364. Defensive / Offensive = over that window the realized price leaned below / above the middle 33%-66% band of the cone (net share of days on that side above 1/3, days beyond the 5%-95% band counted twice). Net driver = sum of the indicator contributions in the forecast decomposition (InHouse model details).

Back-test of the last report dates

MAE weighted: 11.21% | MAE mid: 13.66% | log-loss: 0.8305. Sample: 1,249 daily checks from the last 11 report dates (Nov 2025 to Sep 2026). They cover 216 trading days, because the windows overlap. MAE = mean absolute gap between the path and the realized price, as a share of the price. The most likely zone was right on 60.1% of these checks. Always picking the most common realized zone gives 68.9%.

InHouse vs the banks

Expected path and cone of the InHouse forecast released on 12 Oct 2026 against the forecasts of each bank in the last Consensus Economics survey before it (Sep 2026), one point per month. Five banks are on at the start; click a name in the legend to add or remove it.

InHouse forecast with the banks on top

The InHouse cone and expected path with a smoothed histogram (kernel density) of the bank forecasts at 3, 6, 9 and 12 months, drawn to the right of each date. On the right: the odds and the price of each zone at the end of the series (Offensive: top of the cone, Neutral: expected path, Defensive: bottom of the cone) and the Consensus mean at 12 months, stacked so they do not overlap. The mean of the banks, shaded between the 25% and 75% of them, is off at the start; turn it on from the legend. Pick a month to see that forecast with the banks of its Consensus report (the latest uses the previous report until the new survey is out); past months also show what the price did afterwards.

Forecasts by bank

Pick a report (Consensus Economics surveys since 1995; it opens on the latest). Chart: each bank in grey, from the report date to its last horizon (the long-term forecast is placed 5 years out); purple: mean where every bank answers; blue: the InHouse expected path (12 months) for the report it was compared with; orange: gold from 3 years before the report to its last horizon (or today), so a past report shows what the price did afterwards. Table: USD per ounce at 3, 6, 9 and 12 months and long term (LT), sorted by the 12-month forecast, with an InHouse row for that report.

Consensus Economics asks for quarterly averages, and the long term is the 2031-2035 average. The pipeline dates each point at the end of its period and interpolates the months in between (see Tech debt).