Merge pull request enarjord#174 from 556isback/sharp_gain_ratio

Add Sharp gain ratio in tuning
pbosch · Jul 5, 2021 · 1ef2afc · 1ef2afc
2 parents a49d22d + 438b8f5
commit 1ef2afc
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 3 deletions.
diff --git a/configs/optimize/default.hjson b/configs/optimize/default.hjson
@@ -8,6 +8,8 @@
   # set to 0.0 to disable breaking early
   break_early_factor: 0.5
 
+  minimum_sharpe_ratio: 1.1
+  sharpe_ratio_n_days: 3.0
   minimum_bankruptcy_distance: 0.5
   minimum_equity_balance_ratio: 0.5
   minimum_slice_adg: 1.0

diff --git a/optimize.py b/optimize.py
@@ -107,6 +107,7 @@ def objective_function(analysis: dict, config: dict, metric='adjusted_daily_gain
         * min(1.0, config['maximum_hrs_no_fills_same_side'] / analysis['max_hrs_no_fills_same_side'])
         * min(1.0, analysis['closest_bkr'] / config['minimum_bankruptcy_distance'])
         * min(1.0, analysis['lowest_eqbal_ratio'] / config['minimum_equity_balance_ratio'])
+        * min(1.0, analysis['sharpe_ratio'] / config['minimum_sharpe_ratio'])
     )
 
 
@@ -144,6 +145,7 @@ def single_sliding_window_run(config, data, do_print=False) -> (float, [dict]):
         line = (f'{str(z).rjust(3, " ")} adg {analysis["average_daily_gain"]:.4f}, '
                 f'bkr {analysis["closest_bkr"]:.4f}, '
                 f'eqbal {analysis["lowest_eqbal_ratio"]:.4f} n_days {analysis["n_days"]:.1f}, '
+                f'sharpe_ratio {analysis["sharpe_ratio"]:.4f} , '
                 f'score {analysis["score"]:.4f}, objective {objective:.4f}, '
                 f'hrs stuck ss {str(round(analysis["max_hrs_no_fills_same_side"], 1)).zfill(4)}, ')
         if (bef := config['break_early_factor']) != 0.0:
@@ -163,6 +165,10 @@ def single_sliding_window_run(config, data, do_print=False) -> (float, [dict]):
                 line += f"broke on max_hrs_no_fills_ss {analysis['max_hrs_no_fills_same_side']:.4f}, {config['maximum_hrs_no_fills_same_side']}"
                 print(line)
                 break
+            if analysis['sharpe_ratio'] < config['minimum_sharpe_ratio'] * (1 - bef):
+                line += f"broke on low sharpe ratio {analysis['sharpe_ratio']:.4f} "
+                print(line)
+                break
             if analysis['average_daily_gain'] < config['minimum_slice_adg']:
                 line += f"broke on low adg {analysis['average_daily_gain']:.4f} "
                 print(line)
@@ -188,6 +194,7 @@ def simple_sliding_window_wrap(config, data, do_print=False):
                     daily_gain=np.mean([r['average_daily_gain'] for r in analyses]),
                     closest_bkr=np.min([r['closest_bkr'] for r in analyses]),
                     lowest_eqbal_r=np.min([r['lowest_eqbal_ratio'] for r in analyses]),
+                    sharpe_ratio=np.mean([r['sharpe_ratio'] for r in analyses]),
                     max_hrs_no_fills=np.max([r['max_hrs_no_fills'] for r in analyses]),
                     max_hrs_no_fills_ss=np.max([r['max_hrs_no_fills_same_side'] for r in analyses]))
 
@@ -258,6 +265,7 @@ def backtest_tune(data: np.ndarray, config: dict, current_best: Union[dict, list
             metric_columns=['daily_gain',
                             'closest_bkr',
                             'lowest_eqbal_r',
+                            'sharpe_ratio',
                             'max_hrs_no_fills',
                             'max_hrs_no_fills_ss',
                             'objective'],
@@ -292,8 +300,10 @@ async def main():
         return
     downloader = Downloader(config)
     print()
-    for k in (keys := ['exchange', 'symbol', 'starting_balance', 'start_date', 'end_date', 'latency_simulation_ms',
-                       'do_long', 'do_shrt', 'minimum_bankruptcy_distance', 'maximum_hrs_no_fills',
+    for k in (keys := ['exchange', 'symbol', 'starting_balance', 'start_date',
+                       'end_date', 'latency_simulation_ms',
+                       'do_long', 'do_shrt', 'minimum_sharpe_ratio', 'sharpe_ratio_n_days',
+                       'minimum_bankruptcy_distance', 'maximum_hrs_no_fills',
                        'maximum_hrs_no_fills_same_side', 'iters', 'n_particles', 'sliding_window_days', 'metric',
                        'min_span', 'max_span', 'n_spans']):
         if k in config:

diff --git a/plotting.py b/plotting.py
@@ -19,6 +19,7 @@ def gain_conv(x):
     lines.append(f"symbol {result['symbol'] if 'symbol' in result else 'unknown'}")
     lines.append(f"gain percentage {round_dynamic(result['result']['gain'] * 100 - 100, 4)}%")
     lines.append(f"average_daily_gain percentage {round_dynamic((result['result']['average_daily_gain'] - 1) * 100, 3)}%")
+    lines.append(f"sharpe_ratio percentage {round_dynamic(result['result']['sharpe_ratio'] * 100, 3)}%")
     lines.append(f"closest_bkr percentage {round_dynamic(result['result']['closest_bkr'] * 100, 4)}%")
     lines.append(f"starting balance {round_dynamic(result['starting_balance'], 3)}")
 

diff --git a/pure_funcs.py b/pure_funcs.py
@@ -380,6 +380,7 @@ def get_empty_analysis(bc: dict) -> dict:
         'n_days': 0.0,
         'average_daily_gain': 0.0,
         'adjusted_daily_gain': 0.0,
+        'sharpe_ratio': 0.0,
         'lowest_eqbal_ratio': 0.0,
         'closest_bkr': 1.0,
         'n_fills': 0.0,
@@ -430,7 +431,14 @@ def analyze_fills(fills: list, bc: dict, first_ts: float, last_ts: float) -> (pd
     else:
         shrt_stuck_mean = 0.0
         shrt_stuck = 0.0
-
+
+    ms_span = 1000 * 60 * 60 * 24 * bc['sharpe_ratio_n_days']
+    groups = fdf.groupby(fdf.timestamp // ms_span)
+    periodic_gains = groups.pnl.sum() / groups.balance.first()
+    periodic_gains = periodic_gains.reindex(np.arange(periodic_gains.index[0], periodic_gains.index[-1])).fillna(0.0)
+    periodic_gains_std = periodic_gains.std()
+    sharpe_ratio = periodic_gains.mean() / periodic_gains_std if periodic_gains_std != 0.0 else -20.0
+    sharpe_ratio = np.nan_to_num(sharpe_ratio)
     result = {
         'starting_balance': bc['starting_balance'],
         'final_balance': fdf.iloc[-1].balance,
@@ -440,6 +448,7 @@ def analyze_fills(fills: list, bc: dict, first_ts: float, last_ts: float) -> (pd
         'n_days': (n_days := (last_ts - first_ts) / (1000 * 60 * 60 * 24)),
         'average_daily_gain': (adg := gain ** (1 / n_days) if gain > 0.0 and n_days > 0.0 else 0.0),
         'adjusted_daily_gain': np.tanh(10 * (adg - 1)) + 1,
+        'sharpe_ratio': sharpe_ratio,
         'profit_sum': fdf[fdf.pnl > 0.0].pnl.sum(),
         'loss_sum': fdf[fdf.pnl < 0.0].pnl.sum(),
         'fee_sum': fdf.fee_paid.sum(),