Swiss inflation persistence was detectable in real time
Switzerland's 2021–2023 inflation increase was identifiable as persistent using data available at the time, Simon Beyeler finds in Swiss National Bank research. But the models' trend estimates did not translate into superior inflation forecasts.
Persistence visible by April 2022
Simon Beyeler applies unobserved-components stochastic-volatility models to Swiss consumer price inflation in Swiss National Bank (SNB) Working Paper 14/2026.
One model uses headline inflation; the other uses 22 expenditure groups.
Both include explicit seasonal components because Swiss CPI data are not seasonally adjusted.
Using observations only through April 2022, both attribute most of the post-pandemic increase to the persistent trend, closely matching their full-sample readings.
Their estimated three-month inflation trends reach approximately 0.62 and 0.79 percentage points, respectively, at that cutoff.
Sector-specific trends dominate the sectoral decomposition; the common trend contributes relatively little.
Swiss CPI observations are not revised.
Simple forecasts remain hard to beat
The forecast test covers 180 monthly origins from January 2010 to December 2024, at horizons of one, three, six and twelve months.
One month ahead, a twelve-lag autoregression has an RMSE ratio of 0.509 against a random walk, versus 0.965 and 0.910 for the headline and sectoral models.
A seasonal random walk performs best at six months, with a ratio of 0.624. At twelve months, neither trend model beats the benchmark: ratios reach 1.050 and 1.039. Sectoral detail improves shorter-horizon forecasts only modestly.
Diagnosis is not prediction
The paper supports using these models to assess underlying inflation, not treating them as superior forecasting tools.
Persistence estimates depend heavily on imposed signal-to-noise restrictions.
Much of the intermediate-horizon forecast improvement reflects seasonality rather than better prediction of the trend.