ENH: NPV: Support "gufunc" like behaviour

Author: Kai-Striega

benchmarks/benchmarks.py

@@ -24,8 +24,8 @@ def setup(self, n_cashflows, cashflow_lengths, rates_lengths):
         cf_shape = (n_cashflows, cashflow_lengths)
         self.cashflows = rng.standard_normal(cf_shape)
         self.rates = rng.standard_normal(rates_lengths)
-        self.cashflows_decimal = rng.standard_normal(cf_shape).asdtype(Decimal)
-        self.rates_decimal = rng.standard_normal(rates_lengths).asdtype(Decimal)
+        self.cashflows_decimal = rng.standard_normal(cf_shape).astype(Decimal)
+        self.rates_decimal = rng.standard_normal(rates_lengths).astype(Decimal)
 
     def time_broadcast(self, n_cashflows, cashflow_lengths, rates_lengths):
         npf.npv(self.rates, self.cashflows)

numpy_financial/_financial.py

@@ -46,6 +46,23 @@ def _convert_when(when):
         return [_when_to_num[x] for x in when]
 
 
+def _return_ufunc_like(array):
+    try:
+        # If size of array is one, return scalar
+        return array.item()
+    except ValueError:
+        # Otherwise, return entire array
+        return array
+
+
+def _is_object_array(array):
+    return array.dtype == np.dtype("O")
+
+
+def _use_decimal_dtype(*arrays):
+    return any(_is_object_array(array) for array in arrays)
+
+
 def fv(rate, nper, pmt, pv, when='end'):
     """Compute the future value.
 
@@ -825,6 +842,15 @@ def irr(values, *, guess=None, tol=1e-12, maxiter=100, raise_exceptions=False):
     return np.nan
 
 
+def _to_decimal_array_1d(array):
+    return np.array([Decimal(x) for x in array.tolist()])
+
+
+def _to_decimal_array_2d(array):
+    l = [Decimal(x) for row in array.tolist() for x in row]
+    return np.array(l).reshape(array.shape)
+
+
 def npv(rate, values):
     r"""Return the NPV (Net Present Value) of a cash flow series.
 
@@ -892,15 +918,36 @@ def npv(rate, values):
     3065.22267
 
     """
+    rates = np.atleast_1d(rate)
     values = np.atleast_2d(values)
-    timestep_array = np.arange(0, values.shape[1])
-    npv = (values / (1 + rate) ** timestep_array).sum(axis=1)
-    try:
-        # If size of array is one, return scalar
-        return npv.item()
-    except ValueError:
-        # Otherwise, return entire array
-        return npv
+
+    if rates.ndim != 1:
+        msg = "invalid shape for rates. Rate must be either a scalar or 1d array"
+        raise ValueError(msg)
+
+    if values.ndim != 2:
+        msg = "invalid shape for values. Values must be either a 1d or 2d array"
+        raise ValueError(msg)
+
+    dtype = Decimal if _use_decimal_dtype(rates, values) else np.float64
+
+    if dtype == Decimal:
+        rates = _to_decimal_array_1d(rates)
+        values = _to_decimal_array_2d(values)
+    zero = dtype("0.0")
+    one = dtype("1.0")
+
+    shape = tuple(array.shape[0] for array in (rates, values))
+    out = np.empty(shape=shape, dtype=dtype)
+
+    for i in range(rates.shape[0]):
+        for j in range(values.shape[0]):
+            acc = zero
+            for t in range(values.shape[1]):
+                acc += values[j, t] / ((one + rates[i]) ** t)
+            out[i, j] = acc
+
+    return _return_ufunc_like(out)
 
 
 def mirr(values, finance_rate, reinvest_rate, *, raise_exceptions=False):