Fix CI: study 217 synthetic index overflowed pandas Timestamp bounds

The Lipstick-Index null tape builds 5000 monthly points with
pd.date_range("2000-01-01", periods=5000, freq="MS"), which lands in
2416 — past pandas' ~2262 nanosecond-Timestamp bound — and raised
OutOfBoundsDatetime on the CI runners (3.10–3.12), reddening the whole
tests workflow since the 211-251 batch. It happened to pass on the newer
local pandas only.

The month label is decorative (recession months are drawn at random, not
by date) and the docstring already calls it a period, so switch to
pd.period_range — PeriodIndex has no nanosecond bound. No test reads the
column as a Timestamp; the 217 suite is green.

Author: Guillain-RDCDE

studies/217-lipstick-index/lipstick_index/data.py

@@ -133,7 +133,11 @@ def synthetic_monthly(
     base_m = base_return_ann / 12.0
     vol_m = vol_ann / np.sqrt(12)
 
-    months = pd.date_range("2000-01-01", periods=n_months, freq="MS")
+    # PeriodIndex (not date_range): the label is decorative, and a large n_months
+    # (e.g. the 5000-month null tape) would push a Timestamp index past pandas'
+    # ~2262 nanosecond bound and raise OutOfBoundsDatetime on CI. Periods have no
+    # such bound, and the docstring already calls this column a period.
+    months = pd.period_range("2000-01", periods=n_months, freq="M")
     in_rec = rng.random(n_months) < recession_fraction
 
     spy_rets = rng.normal(base_m, vol_m, n_months)