Interest rate setting by the SNB 2022-24

2024-03

Photo by Phillip Gow on Unsplash

With the SNB’s September 26 meeting more than two months away, there is plenty of time to prepare other tools that may be useful for analysing Swiss monetary policy.

In this post I estimate a simple reaction function for the SNB. Given that the SNB raised interest rates in 2022 after having kept them at -0.75% for 7 years, one may think that there is far too little recent data to estimate a reaction function. Or one may argue that we should do our best with the data that we have. Here I will take the latter view.

While an estimated reaction function is unlikely to be useful for predicting interest rate changes, it can serve as a starting point for thinking about SNB policy. As time passes and more data become available, the model can be updated and refined.

Data

I start by looking at data on the SNB’s policy rate and inflation from 2018 to June 2024. The policy rate was at -0.75% – a world record low – and inflation was also weak until early 2020, when covid struck and inflation turned negative. From early 2021 inflation rose and peaked above 3% in early 2022. It has since subsided and is now well within the SNB’s 0-2% inflation definition of price stability (or inflation objective).  

Source: SNB

In response to these developments, in June 2022 the SNB acted and initiated a series of interest rate increases that took rates to 1.75% by June 2023. It then held rates constant until March 2024, when it cut rates. It cut rates again in June 2024.

A model

Of course, it makes little sense to include in the analysis data from the period during which the policy rate was glued to -0.75%. I therefore start in 2021. Furthermore, the analysis must be limited to the months in which the quarterly SNB meetings took place. That leaves 14 data points. With so few data points, only a very simple model can be fitted.

Below I model the change in the SNB’s policy rate as depending on four variables:

• A constant. This is related to r* in the estimation period. I return to it below.

• The change in interest rates at the last SNB meeting. Central banks often change interest rates several times in the same direction. Thus, we expect the coefficient to be positive but less than unity.

• The rate of inflation. Higher inflation leads the SNB to raise rates.

• The level of the policy rate set at the last meeting. All else equal, if the interest is high (or low), we would expect it to be cut (or raised) over time. Thus, it should enter with a negative sign.  

Estimates

Estimating this equation, we obtain the results in column (1) where t-statistics are shown in brackets, []. All parameters are significant and have the expected sign. It is interesting to note that the parameter on the level of interest rates at the last meeting is of the same size (but of the opposite sign) as the parameter on inflation. That suggests that the SNB raises interest rates when the real interest rate is low, and cuts them when the real interest rate is high.

Imposing that constraint, we obtain the model in the second column. Note that the constant is of the same size (but of the opposite sign) as the parameter on the real interest rate. This tells us something about r*. I introduce that restriction and obtain the model in column (3), which involves two parameters.  

Source: my estimates as described in the text.

The final model has two interesting implications.

1.      The equilibrium real interest rate, r*, was -1% in this sample.¹

2.      If the SNB increased interest rates by 25 bps at one meeting, the expected change in interest rates at the next meeting is increased by 0.54 times 25 bps, or 14 bps.

Fit

To explore the fit of the model, the graph below shows on the horizontal axis the predicted change in the interest rate at each policy meeting and on the vertical axis the actual change. If the model did a perfect job accounting for SNB’s interest rate changes, the data would all fall on a line with a 45-degree angle.

Source: my estimates as described in the text.

The model does a good job in the estimation sample. The correlation between the two series is 0.93 and a fitted line has (almost) a slope of 1 and an intercept of 0.

But the SNB changes interest rates in multiplies of 0.25% and not in basis points. The table below shows the predicted values rounded to the closest 0.25% and the actual interest rate change adopted by the SNB.

Source: my estimates as described in the text.

The model accounts well for the SNB’s interest rate changes in this period. Interestingly, it predicts an interest rate increase two quarters before it occurred. Similarly, it predicts an interest rate cut a quarter before it occurred.

Of course, this finding reflects the fact that I use data for the full period in predicting interest rate changes whereas the SNB only had data until the time of its decisions. Had it known how severe the inflation problem would become, it would have no doubt raised rates earlier.

Conclusions

This analysis shows that a simple model according to which the SNB sets interest rates with reference to its interest rate decision last quarter, the rate of inflation, and which assumes an equilibrium real interest rate of -1 does a good job accounting for SNB’s interest changes in the 2022-24 period.

But all of that is in the estimation period. How well it will predict interest rate changes outside the estimation period is much less clear. SNB Chairman Jordan has already indicated that the SNB now believes that r* has risen to 0. This suggests that this model will underpredict the SNB’s policy rates in the future.

I plan to return to these questions as the next SNB meeting approaches.

As always, I would be keen to hear comments and suggestions for improvements.

The work reported here is preliminary and may be subject to errors. It should not be seen as constituting investment advice. Readers are advised to seek professional investment advice.

1. To see this, consider the steady state in which there are no interest rate changes. In this case, the model can be written as 0 = ‑0.14 – 0.14r* so that r* = -1.