Market Data

The Data Everyone Argues About Is the Data We Can Least Predict

There is an inverse relationship in our forecast record that nobody talks about: the data that dominates the headlines, tariffs, trade flows, fuel prices, is precisely the data we can least reliably forecast. The loudest numbers are the noisiest, and planning around them as if they were solid is a trap.

Grade the forecastability of manufacturing data by category and an uncomfortable pattern falls out: the numbers that generate the most argument are the ones we can least predict. As of Aug 2026, the noisiest category we track is Fuels and spot energy, with a median forecast error of 18.7%, followed by Trade flows at 15%. Compare that to the calmest category, Factory activity, at just 0.5%. The volatile, headline-grabbing series are an order of magnitude harder to forecast than the boring activity series nobody puts on television.

The loud-but-noisy categories

Look at where the error concentrates. Tariff-linked series carry a median error of 8.3%, and trade-flow series 15%, both far above the factory-activity series near 0.5%. This is not a knock on the data; it is the nature of it. Tariffs move on policy decisions, trade flows on front-loading and rerouting, fuels on geopolitics and weather, none of which follows the smooth momentum that makes a series forecastable. The very features that make these numbers newsworthy, their sensitivity to shocks, are the features that make them unpredictable.

Why this inversion matters for planning

The danger is that attention and predictability point in opposite directions. A planning process naturally spends its energy on the numbers in the news, tariffs, trade, fuel, and treats them as inputs it can pin down. But those are exactly the inputs that resist forecasting, so a plan built on confident assumptions about them is built on sand. Meanwhile the genuinely forecastable series, factory activity, hours, wages, get less attention precisely because they are calm. The rational allocation is the reverse of the instinctive one: anchor plans on the predictable series and treat the loud ones as wide scenario ranges, not point estimates.

The numbers that make the news are the ones we can least predict. Attention flows to volatility; reliability hides in the boring series nobody covers.

Plan around what holds still

The takeaway is not to ignore tariffs or fuel, they matter enormously to cost, but to stop pretending you can forecast them and to plan accordingly: hedge them, build wide bands around them, and revisit them often, rather than baking a single confident guess into a model. Reserve the confident assumptions for the series that have earned them with a track record of low error. A plan that knows which of its inputs are solid and which are guesses is far more robust than one that treats every number as equally knowable, especially when the loudest inputs are the least knowable of all.

The noisiest series in the archive

The five-year record shows Brent crude holding a range for years and then breaking it, the pattern most likely to catch anyone carrying a stale assumption. Brent crude drifted through the early part of the archive without a decisive move, then went up 54% in the last two years alone to $114.89, in the upper third of its five-year range. A range that holds that long teaches people to trust it, and that trust is exactly what the break punishes.

The forecast pages break down predictability by category, so you can see which inputs to trust and which to hedge. See the accuracy by category

Published 2026-08-06.