diff --git a/auto_create_timestamps_pr.md b/auto_create_timestamps_pr.md
new file mode 100644
index 0000000000..7dc121ac08
--- /dev/null
+++ b/auto_create_timestamps_pr.md
@@ -0,0 +1,17 @@
+# Auto-Create Timestamps in `prettify_prediction()` When `test_data` is None
+
+## Why are these changes needed?
+
+Currently, the `TimeSeriesDataset.prettify_prediction()` method in `flaml/automl/time_series/ts_data.py` throws a `NotImplementedError` when `test_data` is `None`.
+This is frustrating for users who want to make predictions without providing explicit test data timestamps.
+
+**This PR implements automatic timestamp generation** by:
+
+1. Using the training data's end date as the starting point.
+2. Generating future timestamps based on the inferred frequency.
+3. Supporting `np.ndarray`, `pd.Series`, and `pd.DataFrame`.
+
+## Checks
+
+- [x] Pre-commit linting (black, ruff).
+- [x] Added regression tests demonstrating the fix.
diff --git a/flaml/automl/time_series/ts_data.py b/flaml/automl/time_series/ts_data.py
index 625049c601..e6d872bedb 100644
--- a/flaml/automl/time_series/ts_data.py
+++ b/flaml/automl/time_series/ts_data.py
@@ -243,13 +243,23 @@ def prettify_prediction(self, y_pred: Union[pd.DataFrame, pd.Series, np.ndarray]
                 y_pred[self.time_col] = self.test_data[self.time_col]
 
         else:
+            # Auto-create timestamps when test_data is None
             if isinstance(y_pred, np.ndarray):
-                raise ValueError("Can't enrich np.ndarray as self.test_data is None")
+                y_pred = pd.DataFrame(data=y_pred, columns=self.target_names)
             elif isinstance(y_pred, pd.Series):
                 assert len(self.target_names) == 1, "Not enough columns in y_pred"
                 y_pred = pd.DataFrame({self.target_names[0]: y_pred})
-            # TODO auto-create the timestamps for the time column instead of throwing
-            raise NotImplementedError("Need a non-None test_data for this to work, for now")
+
+            # Generate timestamps based on training data's end_date and frequency
+            train_end_date = self.train_data[self.time_col].max()
+            pred_timestamps = pd.date_range(
+                start=train_end_date,
+                periods=len(y_pred) + 1,
+                freq=self.frequency,
+            )[
+                1:
+            ]  # Skip the first timestamp (train_end_date itself)
+            y_pred[self.time_col] = pred_timestamps
 
         assert isinstance(y_pred, pd.DataFrame)
         assert self.time_col in y_pred.columns
diff --git a/test/automl/test_forecast.py b/test/automl/test_forecast.py
index c77431e167..1f3137f7c2 100644
--- a/test/automl/test_forecast.py
+++ b/test/automl/test_forecast.py
@@ -724,6 +724,124 @@ def test_log_training_metric_ts_models():
             assert automl.best_estimator == estimator
 
 
+def test_prettify_prediction_auto_timestamps_data_types():
+    """Test auto-timestamp generation with different input data types.
+
+    Before this PR fix, calling prettify_prediction() with test_data=None raised:
+    - ValueError for np.ndarray: "Can't enrich np.ndarray as self.test_data is None"
+    - NotImplementedError for pd.Series/DataFrame: "Need a non-None test_data for this to work"
+
+    This test verifies the fix works for np.ndarray, pd.Series, and pd.DataFrame inputs.
+    """
+    from flaml.automl.time_series import TimeSeriesDataset
+
+    # Create training data with daily frequency
+    n = 30
+    train_df = pd.DataFrame(
+        {
+            "date": pd.date_range(start="2023-01-01", periods=n, freq="D"),
+            "value": np.random.randn(n),
+        }
+    )
+    tsds = TimeSeriesDataset(train_df, time_col="date", target_names="value")
+    assert len(tsds.test_data) == 0
+
+    pred_steps = 5
+    expected_start = pd.date_range(start=train_df["date"].max(), periods=2, freq="D")[1]
+
+    # Test np.ndarray
+    result = tsds.prettify_prediction(np.random.randn(pred_steps))
+    assert isinstance(result, pd.DataFrame)
+    assert len(result) == pred_steps
+    assert result["date"].iloc[0] == expected_start
+
+    # Test pd.Series
+    result = tsds.prettify_prediction(pd.Series(np.random.randn(pred_steps)))
+    assert isinstance(result, pd.DataFrame)
+    assert len(result) == pred_steps
+    assert result["date"].iloc[0] == expected_start
+
+    # Test pd.DataFrame
+    result = tsds.prettify_prediction(pd.DataFrame({"value": np.random.randn(pred_steps)}))
+    assert isinstance(result, pd.DataFrame)
+    assert len(result) == pred_steps
+    assert result["date"].iloc[0] == expected_start
+
+
+def test_prettify_prediction_auto_timestamps_frequencies():
+    """Test auto-timestamp generation with different frequencies.
+
+    Before this PR fix, this would raise NotImplementedError when test_data is None.
+    Tests daily and monthly frequencies with np.ndarray input.
+    """
+    from flaml.automl.time_series import TimeSeriesDataset
+
+    pred_steps = 6
+
+    # Test daily frequency
+    train_df_daily = pd.DataFrame(
+        {
+            "date": pd.date_range(start="2023-01-01", periods=30, freq="D"),
+            "value": np.random.randn(30),
+        }
+    )
+    tsds_daily = TimeSeriesDataset(train_df_daily, time_col="date", target_names="value")
+    result = tsds_daily.prettify_prediction(np.random.randn(pred_steps))
+    expected_dates = pd.date_range(start=train_df_daily["date"].max(), periods=pred_steps + 1, freq="D")[1:]
+    pd.testing.assert_index_equal(pd.DatetimeIndex(result["date"]), expected_dates, check_names=False)
+
+    # Test monthly frequency
+    train_df_monthly = pd.DataFrame(
+        {
+            "date": pd.date_range(start="2022-01-01", periods=24, freq="MS"),
+            "value": np.random.randn(24),
+        }
+    )
+    tsds_monthly = TimeSeriesDataset(train_df_monthly, time_col="date", target_names="value")
+    result = tsds_monthly.prettify_prediction(np.random.randn(pred_steps))
+    expected_dates = pd.date_range(start=train_df_monthly["date"].max(), periods=pred_steps + 1, freq="MS")[1:]
+    pd.testing.assert_index_equal(pd.DatetimeIndex(result["date"]), expected_dates, check_names=False)
+
+
+def test_auto_timestamps_e2e(budget=3):
+    """E2E test: train a model and predict without explicit test_data timestamps.
+
+    This showcases the improvement from this PR - users can now make predictions
+    without providing explicit test data timestamps.
+    """
+    try:
+        import statsmodels  # noqa: F401
+    except ImportError:
+        print("statsmodels not installed, skipping E2E test")
+        return
+
+    # Create training data
+    n = 100
+    train_df = pd.DataFrame(
+        {
+            "ds": pd.date_range(start="2020-01-01", periods=n, freq="D"),
+            "y": np.sin(np.linspace(0, 10, n)) + np.random.randn(n) * 0.1,
+        }
+    )
+
+    # Train model
+    automl = AutoML()
+    automl.fit(
+        dataframe=train_df,
+        label="y",
+        period=10,
+        task="ts_forecast",
+        time_budget=budget,
+        estimator_list=["arima"],
+    )
+
+    # Predict using steps (no explicit test_data) - this is the key improvement
+    y_pred = automl.predict(10)
+    assert y_pred is not None
+    assert len(y_pred) == 10
+    print("E2E test passed: model trained and predicted without explicit test_data!")
+
+
 if __name__ == "__main__":
     # test_forecast_automl(60)
     # test_multivariate_forecast_num(5)