Docs: clarify equal-weighted portfolio vs cap-weighted benchmark comp…

docs/component/strategy.rst

@@ -88,6 +88,14 @@ TopkDropoutStrategy
 
 - Generate the order list from the target amount
 
+.. note::
+    `TopkDropoutStrategy` holds an **equal-weighted** portfolio: the available cash is split
+    evenly among the stocks bought on each rebalance. When its results are compared against a
+    capitalization-weighted benchmark such as the default ``SH000300``, the reported excess
+    return also contains the equal- versus cap-weighting spread of the universe, which is a
+    property of the weighting scheme rather than of the signal. Use a benchmark whose weighting
+    scheme matches the strategy if you want to isolate the contribution of the signal.
+
 EnhancedIndexingStrategy
 ------------------------
 `EnhancedIndexingStrategy` Enhanced indexing combines the arts of active management and passive management,

qlib/contrib/strategy/signal_strategy.py

@@ -73,6 +73,24 @@ def get_risk_degree(self, trade_step=None):
 
 
 class TopkDropoutStrategy(BaseSignalStrategy):
+    """
+    An equal-weighted top-k strategy with fixed daily turnover (the ``Topk-Drop`` algorithm).
+
+    .. note::
+        Positions are **equal-weighted**. On each rebalance the available cash is divided
+        evenly among the stocks to be bought (see :meth:`generate_trade_decision`), so the
+        realized portfolio approximates an equal-weighted portfolio of ``topk`` names.
+
+        This is worth keeping in mind when reading the excess return reported by
+        :class:`qlib.workflow.record_temp.PortAnaRecord`, which is a plain difference
+        against a single benchmark instrument. The default benchmark ``SH000300``
+        (CSI 300) is capitalization-weighted, so that difference also contains the return
+        spread between equal- and cap-weighting of the universe, which is a property of
+        the weighting scheme rather than of the signal. To isolate the contribution of the
+        signal, use a benchmark whose weighting scheme matches the strategy, e.g.
+        an equal-weighted portfolio of the same universe run through the same executor.
+    """
+
     # TODO:
     # 1. Supporting leverage the get_range_limit result from the decision
     # 2. Supporting alter_outer_trade_decision

qlib/workflow/record_temp.py

@@ -365,6 +365,20 @@ class PortAnaRecord(ACRecordTemp):
 
         - The return report and detailed positions of the backtest, returned by `qlib/contrib/evaluate.py:backtest`
     - port_analysis.pkl : The risk analysis of your portfolio, returned by `qlib/contrib/evaluate.py:risk_analysis`
+
+    .. note::
+        ``excess_return_without_cost`` and ``excess_return_with_cost`` are computed as a
+        plain arithmetic difference between the portfolio return and the return of the
+        single instrument configured as ``config["backtest"]["benchmark"]``. No adjustment
+        is made for a difference in weighting scheme between the strategy and the benchmark.
+
+        The default configuration pairs an equal-weighted strategy
+        (:class:`~qlib.contrib.strategy.signal_strategy.TopkDropoutStrategy`) with a
+        capitalization-weighted benchmark (``SH000300``). Under that pairing the reported
+        excess return also contains the equal- versus cap-weighting spread of the universe,
+        so it should be read as performance relative to that specific index rather than as
+        the contribution of the signal alone. Choosing a benchmark whose weighting scheme
+        matches the strategy removes that component.
     """
 
     artifact_path = "portfolio_analysis"