Add run_batch method.

Add sharpe ration as internal calc.
Allow weiner path sims to have different sigmas.
Swap ql for pyfeng in notebooks.
Add some features to README.

Author: bsdz

README.md

@@ -2,11 +2,17 @@
 
 Python module for backtesting trading strategies.
 
-Support event driven backtesting, ie `on_open`, `on_close`, etc. Also supports multiple assets.
+Features
 
-Very basic statistics like book cash, mtm and total value. Currently, everything else needs to be deferred to a 3rd party module like `empyrical`.
+* Event driven, ie `on_open`, `on_close`, etc. 
+* Multiple assets.
+* OHLC Asset. Extendable (e.g support additional fields, e.g. Volatility, or entirely different fields, e.g. Barrels per day).
+* Multiple books.
+* Positional and Basket orders. Extendible (e.g. can support stop loss).
+* Batch runs (for optimization).
+* Captures book history including transactions & daily cash, MtM and total values.
 
-There are some basic tests but use at your own peril. It's not production level code.
+The module provides basic statistics like book cash, mtm and total value. Currently, everything else needs to be deferred to a 3rd party module like `empyrical`.
 
 ## Core dependencies
 
@@ -20,7 +26,7 @@ pip install yatbe
 
 ## Usage
 
-Below is an example usage (the performance of the example strategy won't be good).
+Below is an example usage (the economic performance of the example strategy won't be good).
 
 ```python
 import pandas as pd

build.py renamed to build_mypyc.py

@@ -8,12 +8,12 @@ readme = "README.md"
 repository = "https://github.com/bsdz/yabte"
 
 [tool.poetry.build]
-script = "build.py"
+script = "build_mypyc.py"
 generate-setup-file = true
 
 [tool.poetry.dependencies]
 python = "^3.10,<3.13"
-pandas = ">1.5,<3"
+pandas = "^2.2.1"
 scipy = "^1.10.0"
 pandas-stubs = "^2.1.4.231227"
 mypy = "^1.8.0"
@@ -48,6 +48,7 @@ plotly = "^5.10.0"
 ipykernel = "^6.20.2"
 pyfeng = "^0.2.5"
 nbconvert = "^7.2.9"
+quantlib = "^1.34"
 
 [tool.isort]
 profile = "black"

poetry.lock

@@ -1 +1,4 @@
-_author__ = "Blair Azzopardi"
+from importlib.metadata import version
+
+__author__ = "Blair Azzopardi"
+__version__ = version(__package__)

pyproject.toml

@@ -1,7 +1,9 @@
+import concurrent.futures
 import logging
+from concurrent.futures import ProcessPoolExecutor
 from copy import deepcopy
 from dataclasses import dataclass, field
-from typing import Any, Dict, List, Optional
+from typing import Any, Dict, Iterable, List, Optional
 
 import pandas as pd
 
@@ -219,3 +221,14 @@ def run(self, params: Dict[str, Any] = None) -> StrategyRunnerResult:
                 book.eod_tasks(ts, day_data, asset_map)
 
         return srr
+
+    def run_batch(
+        self,
+        params_iterable: Iterable[Dict[str, Any]],
+        executor: ProcessPoolExecutor | None = None,
+    ) -> List[StrategyRunnerResult]:
+        """Run a set of parameter combinations."""
+
+        executor = executor or concurrent.futures.ThreadPoolExecutor()
+        with executor:
+            return list(executor.map(self.run, params_iterable))

yabte/__init__.py

@@ -5,6 +5,7 @@
 import numpy as np
 import pandas as pd
 
+import yabte.utilities.pandas_extension
 from yabte.backtest import (
     BasketOrder,
     Book,
@@ -571,6 +572,34 @@ def on_close(self):
             [(o.status, o.key, o.size) for o in srr.orders_unprocessed],
         )
 
+    def test_run_batch(self):
+
+        book = Book(name="Main", cash=Decimal("100000"))
+
+        sr = StrategyRunner(
+            data=self.df_combined,
+            assets=self.assets,
+            strategies=[TestSMAXOStrat()],
+            books=[book],
+        )
+
+        param_iter = [
+            {"days_long": n, "days_short": m}
+            for n, m in zip([20, 30, 40, 50], [5, 10, 15, 20])
+            if n > m
+        ]
+
+        srrs = sr.run_batch(param_iter)
+
+        self.assertEqual(len(srrs), len(param_iter))
+
+        # check we have distinct sharpe ratios for each param set
+        sharpes = {
+            srr.book_history.loc[:, ("Main", "total")].prc.sharpe_ratio()
+            for srr in srrs
+        }
+        self.assertEqual(len(sharpes), len(param_iter))
+
 
 if __name__ == "__main__":
     unittest.main()

yabte/backtest/strategyrunner.py

@@ -13,22 +13,23 @@ def standard(self):
 
 
 @pd.api.extensions.register_dataframe_accessor("prc")
+@pd.api.extensions.register_series_accessor("prc")
 class PriceAccessor:
     # TODO: add ledoit cov (via sklearn)
     # http://www.ledoit.net/honey.pdf
-    # TODO: add Sharpe ratio
 
     def __init__(self, pandas_obj):
         self._validate(pandas_obj)
         self._obj = pandas_obj
 
     @staticmethod
     def _validate(obj):
-        if (obj < 0).any(axis=None):
-            raise AttributeError("Prices must be non-negative")
+        pass
 
     @property
     def log_returns(self):
+        if (self._obj < 0).any(axis=None):
+            raise AttributeError("Prices must be non-negative for log returns")
         return np.log((self._obj / self._obj.shift())[1:])
 
     @property
@@ -42,6 +43,8 @@ def frequency(self):
             return 252
         elif days == 7:
             return 52
+        elif 28 <= days <= 31:
+            return 12
 
     def capm_returns(self, risk_free_rate=0):
         returns = self.returns
@@ -56,6 +59,11 @@ def capm_returns(self, risk_free_rate=0):
             + betas * (returns_mkt.mean() * self.frequency - risk_free_rate)
         ).rename("CAPM")
 
+    def sharpe_ratio(self, risk_free_rate=0, use_log_returns=True):
+        ann_factor = np.sqrt(self.frequency)
+        returns = self.log_returns if use_log_returns else self.returns
+        return ann_factor * (returns.mean() - risk_free_rate) / returns.std()
+
     def null_blips(self, sd=5, sdd=7):
         df = self._obj
         z = df.scl.standard

yabte/tests/test_strategy_runner.py

@@ -14,18 +14,20 @@
 def weiner_simulate_paths(
     n_steps: int,
     n_sims: int = 1,
-    stdev: float = 1,
+    stdev: float | np.ndarray = 1,
     R: np.ndarray = np.array([[1]]),
     rng=None,
 ):
     """Generate simulated Weiner paths.
 
     `stdev` is the increment size, `R` a correlation matrix, `n_steps`
     is how many time steps, `n_sims` the number of simulations and `rng`
-    a numpy random number generator (optional).
+    a numpy random number generator (optional). If `stdev` is a scalar
+    it will be broadcasted to the size of `n_sims`.
     """
 
     R = np.atleast_2d(R)
+    stdev = np.resize(stdev, n_sims).reshape(n_sims, 1)
 
     if rng is None:
         rng = np.random.default_rng()