Fixed tests are written incorrectly: user function should be called per test block, and not in batch in advance, because user logs will cluster together before any test blocks, which cannot be correlated to the specific test assertion (and is pretty unreadable too).

Also, tests are not displaying expected value (0 is not the expected balance.), which is unacceptable.

ALSO, failure assertion message for the random tests is wrong: every one of them displays Transactions: [('2015-01-01 10:12:13', 123.45), ('2015-01-01 15:12:13', 123.45), ('2015-02-07 18:00:00', 123.45)] but this is the first fixed test, not the actual transactions being tested.

The third parameter should be renamed to 'initial_balance'.

I added random tests now, thanks for your help.

It's common to put the reference solution in the complete test.
If you are considering about anti-cheat, firstly, you may have to think of what to prevent.

You are right, I wanted to add some random tests but haven't yet.
Is there a better way to access my solution to use it in the tests to calculate the result for the random input than copying it to the test code under a random name?

Needs Random Tests

If the balance is negative, let's say -100 for 10 days and interest per year (assuming 365 for this reply) is 0.01 or 1%, then you would calculate -100 * 10 days * (0.01 / 365) = -0,027397260273973
So, yes it would be deducted, you would have to pay interest to the bank for that period.

If the balance becomes negative over a period of the year, the same interest rate is still used (no penalty is applied)

So should interest rate stay the same or be deducted?