Tool · Updated monthly

Swiss Confederation bonds, FRP 4 and inflation: four-year projections

Every month we project the 10-year Swiss Confederation bond yield, the FRP 4 upper limit of the technical interest rate and Swiss inflation with a neural network autoregression (NNAR) model. The bands shown are statistical scenarios; they are not a forecast by allea.

Observed data to end of August 2026 · projection computed on 10 September 2026

Confederation 10-year

0.47%

August 2026

FRP 4 upper limit

2.85%

in force from 1 October 2025 · next reset on 1 October 2026

Inflation CPI, year on year

0.35%

July 2026

10-year Swiss Confederation bond yield History since 2009 · 45-month projection

10-year spot rate of the Swiss Confederation bond curve (SNB, Nelson-Siegel-Svensson method), month-end. Bands: 95%, 90%, 60%, 30% and 4% intervals of the simulated paths; solid line: median. Gold dashed line: 10-year forward rate implied by the SNB curve of the reference month, a market benchmark independent of the model.

10-year Swiss Confederation bonds, spot rate in % — quantiles of NNAR paths; source: SNB, calculations allea Ltd
Horizon	5%	Median	95%	Market forward
Last observed value (August 2026)	–	0.47	–	–
end of 2026	-0.07	0.42	0.93	0.49
end of 2027	-0.65	0.38	1.45	0.57
end of 2028	-1.06	0.33	1.77	0.63
end of 2029	-1.38	0.29	1.99	0.67

Upper limit of the technical interest rate under FRP 4 Published values since 2009 · projection to 2030

Set each 30 September by the Swiss Chamber of Pension Actuaries (SKPE), applicable from 1 October. Shaded zone: from the next reset.

How is the upper limit calculated?

Technical guideline 4 (FRP 4) of the Swiss Chamber of Pension Actuaries sets the upper limit of the technical interest rate at the average of the last twelve monthly 10-year Swiss Confederation spot rates plus 2.5 percentage points, for generational mortality tables. For period tables, 0.30 percentage points are deducted. The limit is capped at 4.5%. The value in force since 1 October 2025 is 2.85% (source: SKPE press release of 30 September 2025).

Because the formula depends only on the last twelve 10-year rates, the projection of the upper limit follows directly from the previous chart: each simulated path of the 10-year rate yields a path of the FRP 4 limit, whose quantiles we report at each 1 October. The table also gives the model’s probability that the published limit will be below or above the value in force, and the value it would take if rates followed the market forward curve.

FRP 4 upper limit of the technical interest rate, in % — source: SKPE; projections allea Ltd
Horizon	5%	Median	95%	P(lower)	P(higher)	Market forward
Last observed value (August 2026)	–	2.85	–	–	–	–
from 1 October 2026	2.83	2.85	2.87	33%	31%	2.85
from 1 October 2027	2.32	2.91	3.53	43%	56%	3.02
from 1 October 2028	1.78	2.86	4.00	49%	50%	3.08
from 1 October 2029	1.38	2.82	4.29	51%	48%	3.14

Swiss inflation: CPI, year on year History since 1992 · 48-month projection

Swiss consumer price index (FSO), change versus the same month of the previous year.

Swiss inflation (CPI, year on year), in % — source: FSO via SNB; projections allea Ltd
Horizon	5%	Median	95%
Last observed value (July 2026)	–	0.35	–
end of 2026	0.09	1.34	2.62
end of 2027	-1.17	0.97	3.18
end of 2028	-1.24	0.93	3.15
end of 2029	-1.26	0.89	3.14

Method and sources

An NNAR model (neural network autoregression) is a neural network with one hidden layer whose inputs are the past values of the series. For the 10-year yield it is estimated on the monthly changes of the yield, with no trend imposed; for inflation, on the monthly changes of the price index, which brings in the base effects already known for the next twelve months. It is re-estimated every month on the full history since 1994, then simulated over 20,000 paths by adding at each step a random term drawn from the estimation residuals.

The coloured bands are the quantiles of these paths: the lightest contains 95% of them, the darkest 4% around the median. The further the horizon, the wider the range: this is the direct expression of the model’s uncertainty.

Has the model been checked?

Yes, with a rolling backtest: since January 2008 the projection is recomputed every month using only the data available at the time, then compared with what actually happened. The table summarises the result for the published specification. A well-calibrated 90% band should contain about 90% of the realised values; a CRPS below 1 means the projected distribution is more informative than a random walk. The projections published here since September 2026 are kept and will be compared, month after month, with the observed values.

Rolling backtest January 2008 – July 2026: projections recomputed each month as they would have been, compared with realised values
Horizon (months)	90% band: share of realised values covered	Mean absolute error of the median (pp)	CRPS relative to a random walk
Confederation 10-year
6	90%	0.28	0.99
12	91%	0.44	0.98
24	92%	0.65	0.94
36	95%	0.68	0.90
CPI inflation
6	91%	0.55	0.83
12	89%	0.81	0.80
24	90%	0.89	0.68
36	89%	0.93	0.69
FRP 4 limit, 12 months before the reset
12	94%	0.26	–

Sources: Swiss National Bank (parameters of the Swiss Confederation bond yield curve), Federal Statistical Office (consumer price index), SKPE (technical guideline 4 and annual press releases). Calculations: allea Ltd.

What is FRP 4?

Technical guideline 4 of the Swiss Chamber of Pension Actuaries governs the recommendation of the technical interest rate by the pension fund expert. It defines an upper limit that the recommended technical interest rate must not exceed without justification.

Why a range rather than a single value?

A single value would hide the uncertainty. The range shows the spread of paths consistent with the history; its width is as informative as its median.

How often are the projections updated?

Every month, as soon as the SNB has published the previous month’s data. The computation date is shown at the top of the page; earlier results remain available in the archive below.

These projections are statistical scenarios produced by a model. They are neither a forecast by allea, nor a recommendation, nor a guarantee: actual future values may differ materially. allea Ltd accepts no liability for their use.

Downloads

Monthly quantiles (CSV) and charts (PNG, dark and light themes) of the latest computation. Earlier computations are kept in the archive.

nnar-quantiles-latest.csv64 KB
nnar-quantiles-2026-09.csv64 KB

Would you like to discuss the technical interest rate of your pension fund?

Write to info@allea.ch

+41 44 287 80 80