[kritagyadubey]

Fixes #166

Problem

fv sizes and types its output buffer from rate alone (np.empty_like(rate)), so an integer rate gives an integer buffer and the float result is truncated toward zero: npf.fv(0, 12, -100.5, -1000.25) returns 2206 instead of 2206.25.

ipmt and ppmt are affected through _rbl, which calls fv. The remaining-balance array comes back integer typed, and _value_like cannot store NaN in it, so they raise ValueError: cannot convert float NaN to integer before any arithmetic happens:

>>> npf.ipmt(0, 1, 12, 1200)
ValueError: cannot convert float NaN to integer
>>> npf.ppmt(0, 1, 12, 1200)
ValueError: cannot convert float NaN to integer

This only takes an extra float (0.0 instead of 0) to avoid, which is easy to miss for the common case of a 0% loan.

Change

Derive the dtype from all the inputs when allocating the fv output buffer:

fv_array = np.empty_like(rate, dtype=np.result_type(rate, nper, pmt, pv, 1.0))

This matches the promotion the rest of the module already gets from arithmetic. pmt, ppmt and pv all return float64 for the same integer inputs, so fv was the only function affected. Inputs that are already float are unchanged, since result_type treats the 1.0 literal as a weak scalar, and Decimal inputs still produce an object array.

Testing

Run on Python 3.14.7, numpy 2.5.4, Windows:

• spin test -- --doctest-modules: 122 passed, including three new regression tests (TestFv::test_gh_166, TestIpmt::test_gh_166, TestPpmt::test_gh_166). All three fail without the change to _financial.py.
• ruff check numpy_financial/ benchmarks/, mypy --no-incremental . and pyright (the checks in spin lint): no findings.