Case studies
Three real backtest results (see Backtest for the full set), reframed as short situation/so-what/recommendation memos instead of the trader-style write-up used on Notable Calls — a different narrative template over the same real data, not new analysis.
All three are backtested — retrospective runs of the mechanical trend model against real historical data — not live, real-time-published predictions.
UK GDP growth, 2022
2022 sat inside the UK's post-pandemic recovery window, where growth rates were swinging far outside their pre-2020 norms. A naive trend model fit on the preceding real GDP growth history — which includes the 2020 collapse and 2021 rebound — produced an unusually wide ±3.89pp band around its point estimate.
The wide band still wasn't wide enough. Actual growth of +5.15% came in just above the model's own upper bound of +4.34% — a genuine miss, but a narrow one, and a different kind of miss than blowing through a narrow band entirely (compare the Germany case below). The model correctly read real historical volatility into an unusually wide range; the real recovery was still stronger than even that.
An honestly wide band, built from real volatility, isn't a guarantee — it lowers the odds of a miss, it doesn't remove them. Read how far past the edge an actual lands, not just hit-or-miss: this one cleared the bound by 0.81pp, not by several multiples of it.
US inflation, 2020
2020 was a structural-break year: COVID-era demand collapse, followed by the supply-driven inflation surge that didn't peak until 2022. A trend model fit only on pre-2020 history has no way to know a break is coming.
It hit anyway — 1.23% actual landed comfortably inside the stated range — but that's partly luck of timing: the demand shock hadn't yet been offset by the stimulus-driven surge that would blow well past this same model's band in 2021-22. A model that's right for reasons it can't take credit for is still a model whose limits need stating plainly.
Don't read one hit as validation through a structural break — read the following year instead. This is precisely the scenario the site's own miss-cause taxonomy exists for (a real 'structural break' category, not an excuse invented after the fact); a mechanical model earning credit here would be the wrong lesson to take from it.
Germany inflation, 2022
Germany's inflation history through 2021 was famously low and stable — decades of ECB-anchored, sub-3% prints. A mechanical trend model fit on that real history naturally produced a narrow, low band: it had no genuinely volatile regime in its training data to widen it.
The 2022 energy-price shock that followed Russia's invasion of Ukraine pushed German inflation to its highest level in decades — a structural break the model, built only on pre-2022 history, had no way to see coming. The actual print landed nearly 4 percentage points above the model's upper bound, one of the largest misses in the whole backtest set.
A narrow band built on a genuinely stable history isn't a stronger prediction — it's a prediction with more room to be wrong once the regime actually shifts. This is exactly why the backtest shows misses as prominently as hits: a track record built only from calm periods would look artificially strong right up until it wasn't.