Markets

Every day, our server looks at the last three years of closing prices from our own price archive — the same data behind the charts above. From those prices it measures two things: the metal’s recent drift (has it been trending up or down lately, with more weight on the most recent months) and its volatility (how big its typical day-to-day swings are, again weighted toward recent behavior — the same “EWMA” technique banks use for risk management).

Then, instead of drawing one guess, it plays the next 30 trading days out 20,000 times — a Monte Carlo simulation. Each simulated future takes a random daily step consistent with the measured drift and volatility. Some simulated futures drift higher, some lower, most stay in the middle.

The shaded fan is simply where those 20,000 futures ended up: the darkest core contains half of them (the 25th–75th percentile), each lighter ring contains more, and the palest edge contains 90% of them (the 5th–95th percentile). The dashed line is the median — half the simulations finished above it, half below. The further out in time, the wider the fan: uncertainty grows with distance, and the chart is honest about that.

This way of showing a forecast is called a fan chart. It was introduced by the Bank of England in 1996 for its inflation projections and is now the standard used by central banks worldwide — you may know its cousin, the hurricane “cone of uncertainty.” The computation runs nightly in Julia, a language built for scientific computing, on our own servers. It knows nothing about news, policy, or supply — only price history — so treat it as a measure of what recent behavior implies, never as a promise.