It is not so clear if the pillar 3a is strictly bad from a purely payoff point of view when not earning enough.
Let’s say we earn some money and want to invest it. We have two options that will be the same except the account: 3a and regular. As an example, I think investing in the SPI at 0.1% TER with Truewealth 3a (UBS (CH) Index Fund - Equities Switzerland All NSL I-A-acc, CH0348228609) vs IBKR (iShares Core SPI (CH), CH0237935652) should be comparable.
Ok, what are our returns for both solutions?
where:
- r are returns (e.g. a growth of 1%). I also split the yearly returns into capital gains and income.
- t are tax rates (e.g. a tax of 10%). Note, this is the marginal tax bracket for the reg (-ular) taxes.
- y are the number of years that we compound
We now make two simplifying changes:
- We replace the compounding product with an assumption of average returns
- We use the commutativity of the multiplication to pull the 3a exit tax to the front
We see that we can use the regular account to only pay the tax rate of year 1 instead of the 3a tax rate in year y, but then we will pay taxes on all dividends in all years.
Now, let’s assume our tax rate in year 1 is very favorable, it is 0%. Let’s assume we will pay a roughly average 19% marginal taxes at a roughly average 80k income later. We assume 5% capital gains and 2% dividends. We will let it compound for 40 years.
We now want to know:
So, in our example, if our 3a exit tax will be below 13%, we would do 3a even if we would pay no taxes this year. Only in the canton of Zurich, it can even reach 13% (under current tax law).
Rerunning this with different numbers will yield slightly different results. But the main point is this:
The 3a exit tax is also significantly competing with the recurring income taxes on dividend & interest over the compounding period, not only with the initial tax savings.
Also:
The taxation of capital gains in 3a is an illusion when looking at total returns.
Edit: I totally forgot wealth taxes, which strictly pushes the calculus even further in favor of 3a (but would bloat the model).