Add tilt and timing performance attribution. Also respect time range (#111)

twiecki · web-flow · commit a9e6a7172b27 · 2019-08-15T14:37:59.000+02:00
* ENH Add tilt and timing performance attribution. Also respect time ranges.
diff --git a/empyrical/perf_attrib.py b/empyrical/perf_attrib.py
@@ -1,3 +1,4 @@
+from collections import OrderedDict
 import pandas as pd
 
 
@@ -81,17 +82,36 @@ def perf_attrib(returns,
     ----
     See https://en.wikipedia.org/wiki/Performance_attribution for more details.
     """
+
+    # Make risk data match time range of returns
+    start = returns.index[0]
+    end = returns.index[-1]
+    factor_returns = factor_returns.loc[start:end]
+    factor_loadings = factor_loadings.loc[start:end]
+
+    factor_loadings.index = factor_loadings.index.set_names(['dt', 'ticker'])
+
+    positions = positions.copy()
+    positions.index = positions.index.set_names(['dt', 'ticker'])
+
     risk_exposures_portfolio = compute_exposures(positions,
                                                  factor_loadings)
 
     perf_attrib_by_factor = risk_exposures_portfolio.multiply(factor_returns)
-
     common_returns = perf_attrib_by_factor.sum(axis='columns')
+
+    tilt_exposure = risk_exposures_portfolio.mean()
+    tilt_returns = factor_returns.multiply(tilt_exposure).sum(axis='columns')
+    timing_returns = common_returns - tilt_returns
     specific_returns = returns - common_returns
 
-    returns_df = pd.DataFrame({'total_returns': returns,
-                               'common_returns': common_returns,
-                               'specific_returns': specific_returns})
+    returns_df = pd.DataFrame(OrderedDict([
+        ('total_returns', returns),
+        ('common_returns', common_returns),
+        ('specific_returns', specific_returns),
+        ('tilt_returns', tilt_returns),
+        ('timing_returns', timing_returns)
+        ]))
 
     return (risk_exposures_portfolio,
             pd.concat([perf_attrib_by_factor, returns_df], axis='columns'))
diff --git a/empyrical/tests/test_perf_attrib.py b/empyrical/tests/test_perf_attrib.py
@@ -42,12 +42,15 @@ def test_perf_attrib_simple(self):
 
         expected_perf_attrib_output = pd.DataFrame(
             index=dts,
-            columns=['risk_factor1', 'risk_factor2', 'common_returns',
-                     'specific_returns', 'total_returns'],
+            columns=['risk_factor1', 'risk_factor2', 'total_returns',
+                     'common_returns', 'specific_returns',
+                     'tilt_returns', 'timing_returns'],
             data={'risk_factor1': [0.025, 0.025],
                   'risk_factor2': [0.025, 0.025],
                   'common_returns': [0.05, 0.05],
                   'specific_returns': [0.05, 0.05],
+                  'tilt_returns': [0.05, 0.05],
+                  'timing_returns': [0.0, 0.0],
                   'total_returns': returns}
         )
 
@@ -79,12 +82,15 @@ def test_perf_attrib_simple(self):
 
         expected_perf_attrib_output = pd.DataFrame(
             index=dts,
-            columns=['risk_factor1', 'risk_factor2', 'common_returns',
-                     'specific_returns', 'total_returns'],
+            columns=['risk_factor1', 'risk_factor2', 'total_returns',
+                     'common_returns', 'specific_returns',
+                     'tilt_returns', 'timing_returns'],
             data={'risk_factor1': [0.0, 0.0],
                   'risk_factor2': [0.0, 0.0],
                   'common_returns': [0.0, 0.0],
                   'specific_returns': [0.1, 0.1],
+                  'tilt_returns': [0.0, 0.0],
+                  'timing_returns': [0.0, 0.0],
                   'total_returns': returns}
         )
 
@@ -101,6 +107,41 @@ def test_perf_attrib_simple(self):
         pd.util.testing.assert_frame_equal(expected_exposures_portfolio,
                                            exposures_portfolio)
 
+        # test long and short positions with tilt exposure
+        positions = pd.Series([1.0, -0.5, 1.0, -0.5], index=index)
+
+        exposures_portfolio, perf_attrib_output = perf_attrib(returns,
+                                                              positions,
+                                                              factor_returns,
+                                                              factor_loadings)
+
+        expected_perf_attrib_output = pd.DataFrame(
+            index=dts,
+            columns=['risk_factor1', 'risk_factor2', 'total_returns',
+                     'common_returns', 'specific_returns',
+                     'tilt_returns', 'timing_returns'],
+            data={'risk_factor1': [0.0125, 0.0125],
+                  'risk_factor2': [0.0125, 0.0125],
+                  'common_returns': [0.025, 0.025],
+                  'specific_returns': [0.075, 0.075],
+                  'tilt_returns': [0.025, 0.025],
+                  'timing_returns': [0.0, 0.0],
+                  'total_returns': returns}
+        )
+
+        expected_exposures_portfolio = pd.DataFrame(
+            index=dts,
+            columns=['risk_factor1', 'risk_factor2'],
+            data={'risk_factor1': [0.125, 0.125],
+                  'risk_factor2': [0.125, 0.125]}
+        )
+
+        pd.util.testing.assert_frame_equal(expected_perf_attrib_output,
+                                           perf_attrib_output)
+
+        pd.util.testing.assert_frame_equal(expected_exposures_portfolio,
+                                           exposures_portfolio)
+
     def test_perf_attrib_regression(self):
 
         positions = pd.read_csv('empyrical/tests/test_data/positions.csv',
