diff --git a/src/quantlib_st/cli/corr_cmd.py b/src/quantlib_st/cli/corr_cmd.py
index 1d085b7..5e9b80e 100644
--- a/src/quantlib_st/cli/corr_cmd.py
+++ b/src/quantlib_st/cli/corr_cmd.py
@@ -85,6 +85,15 @@ def add_corr_subcommand(subparsers: argparse._SubParsersAction) -> None:
         default=False,
         help="If true, treat input as prices. If false (default), treat as returns.",
     )
+    parser.add_argument(
+        "--signed-log-transform",
+        action=argparse.BooleanOptionalAction,
+        default=True,
+        help=(
+            "If true (default), apply signed-log transform when --is-price-series "
+            "to allow non-positive prices."
+        ),
+    )
 
     parser.add_argument(
         "--index-col",
@@ -121,6 +130,7 @@ def run_corr(args: argparse.Namespace) -> int:
         frequency=args.frequency,
         forward_fill_price_index=args.forward_fill_price_index,
         is_price_series=args.is_price_series,
+        signed_log_transform=args.signed_log_transform,
         date_method=args.date_method,
         rollyears=args.rollyears,
         interval_frequency=args.interval_frequency,
diff --git a/src/quantlib_st/correlation/README.md b/src/quantlib_st/correlation/README.md
index 37f171b..c299c7e 100644
--- a/src/quantlib_st/correlation/README.md
+++ b/src/quantlib_st/correlation/README.md
@@ -19,6 +19,9 @@ cat sample_data/returns_10x4.csv | ./dist/quantlib corr
 - `--interval-frequency 12M`: Controls how often a new correlation matrix is emitted.
 - `--ew-lookback 250` and `--min-periods 20`: Parameters for EWMA estimation.
 - `--is-price-series`: Treat input data as price series and convert to lognormal returns.
+- `--signed-log-transform / --no-signed-log-transform`: When using `--is-price-series`,
+  apply a signed-log transform ($\mathrm{sign}(x)\cdot\log(1+|x|)$) to allow
+  non-positive prices (default: enabled).
 
 ### Key concepts
 
diff --git a/src/quantlib_st/correlation/correlation_over_time.py b/src/quantlib_st/correlation/correlation_over_time.py
index 6cabd2a..521bf68 100644
--- a/src/quantlib_st/correlation/correlation_over_time.py
+++ b/src/quantlib_st/correlation/correlation_over_time.py
@@ -60,6 +60,7 @@ def correlation_over_time(
     frequency: str = "D",
     forward_fill_price_index: bool = True,
     is_price_series: bool = False,
+    signed_log_transform: bool = True,
     **kwargs,
 ) -> CorrelationList:
     """Construct a CorrelationList from a DataFrame of returns or prices.
@@ -86,17 +87,29 @@ def correlation_over_time(
     is_price_series : bool, optional
         If True, treat input as a price index and log-transform it. If False
         (default), treat input as log-normal returns.
+    signed_log_transform : bool, optional
+        If True (default), apply a signed-log transform when ``is_price_series``
+        to allow non-positive prices (e.g., spreads). If False, use ``np.log``
+        which requires strictly positive prices.
     **kwargs :
         All other keyword arguments are forwarded to :func:`compute_correlation_over_time`.
     """
     # 1. Determine the (log) Price Index
     if is_price_series:
         # Input is Price P; we want ln(P) so that diff() yields log-returns.
-        index_prices_for_correlation = pd.DataFrame(
-            np.log(data_for_correlation),
-            index=data_for_correlation.index,
-            columns=data_for_correlation.columns,
-        )
+        if signed_log_transform:
+            # Allow non-positive prices: sign(x) * log(1 + |x|).
+            index_prices_for_correlation = pd.DataFrame(
+                np.sign(data_for_correlation) * np.log1p(np.abs(data_for_correlation)),
+                index=data_for_correlation.index,
+                columns=data_for_correlation.columns,
+            )
+        else:
+            index_prices_for_correlation = pd.DataFrame(
+                np.log(data_for_correlation),
+                index=data_for_correlation.index,
+                columns=data_for_correlation.columns,
+            )
     else:
         # Input is log-normal returns; cumsum yields the log-price index.
         index_prices_for_correlation = data_for_correlation.cumsum()