fix: handle case where informative does not overlap with the main data

Author: xmatthias

freqtrade/strategy/strategy_helper.py

@@ -98,14 +98,15 @@ def merge_informative_pair(
             # We can fill these with the last available informative candle before the start date
             # while still avoiding lookahead bias - as only past data is used.
             first_valid_idx = dataframe[date_merge].first_valid_index()
-            first_valid_date_merge = dataframe.at[first_valid_idx, date_merge]
-            matching_informative_raws = informative[
-                informative[date_merge] < first_valid_date_merge
-            ]
-            if not matching_informative_raws.empty:
-                dataframe.loc[: first_valid_idx - 1] = dataframe.loc[: first_valid_idx - 1].fillna(
-                    matching_informative_raws.iloc[-1]
-                )
+            if first_valid_idx:
+                first_valid_date_merge = dataframe.at[first_valid_idx, date_merge]
+                matching_informative_raws = informative[
+                    informative[date_merge] < first_valid_date_merge
+                ]
+                if not matching_informative_raws.empty:
+                    dataframe.loc[: first_valid_idx - 1] = dataframe.loc[
+                        : first_valid_idx - 1
+                    ].fillna(matching_informative_raws.iloc[-1])
     else:
         dataframe = pd.merge(
             dataframe, informative, left_on="date", right_on=date_merge, how="left"