Updated backtesting with timer.

.improvements

@@ -0,0 +1,2 @@
+1. Model selector based on LOOCV.
+2. Add normal copula model (student's t or normal).

backtest.py

@@ -1,6 +1,5 @@
 # -*- coding: utf-8 -*-
-import torch
-import numpy as np
+import time
 import pandas as pd
 
 from tests.test_config import TEST_YAHOO_STOCK_UNIVERSE_16, TEST_YAHOO_STOCK_UNIVERSE_8, TEST_YAHOO_STOCK_UNIVERSE_4
@@ -10,7 +9,7 @@
 from src.abacus.optimizer.optimizer import MPCMaximumReturn
 
 
-
+# Backtesting setup.
 instrument_specification = {}
 inital_weights = {}
 wealth = []
@@ -20,25 +19,34 @@
     time_series = pd.read_csv(file, index_col='Date')
     instrument_specification[ticker] = time_series
     if i < number_of_start_assets: inital_weights[ticker] = 1 / number_of_start_assets
-
 universe = Universe(instrument_specifications=instrument_specification)
 portfolio = Portfolio(weights=inital_weights)
-simulator = Simulator(universe)
-simulator.calibrate()
-simulator.run_simulation(time_steps=5, number_of_simulations=100)
 
 # Date range for backtesting.
 start_date = "2020-01-02"
-end_date = "2020-01-03" # "2023-05-31"
+end_date = "2020-01-05" # "2023-05-31"
 date_range = pd.date_range(start=start_date, end=end_date, freq='B')
+solutions = {}
+times = {}
 
 for date in date_range:
+    t1 = time.time()
     universe.date_today = date
     simulator = Simulator(universe)
     simulator.calibrate()
-    simulator.run_simulation(time_steps=5, number_of_simulations=25)
-    optimizer = MPCMaximumReturn(universe, portfolio, simulator.return_tensor, gamma=1, l1_penalty=0, l2_penalty=0.05, covariance_matrix=simulator.covariance_matrix)
+    simulator.run_simulation(time_steps=10, number_of_simulations=1000)
+    optimizer = MPCMaximumReturn(universe, portfolio, simulator.return_tensor, gamma=1, l1_penalty=0, l2_penalty=0.05,
+                                 covariance_matrix=simulator.covariance_matrix)
     optimizer.solve()
 
-    print(optimizer.solution)
-    break
+    solution = optimizer.solution
+    times[date] = time.time() - t1
+    portfolio.weights = solution
+    solutions[date] = solution
+
+print('\n' * 10)
+for date, solution in solutions.items():
+    print(f"Took {times[date]} seconds.")
+    for a, w in solution.items():
+        print(a, w)
+    print()

src/abacus/config.py

@@ -6,7 +6,7 @@
 # Processes
 MAXIMUM_AR_ORDER = 5
 INITIAL_VARIANCE_GARCH_OBSERVATIONS = 20
-INITIAL_GARCH_PARAMETERS = (0.05, 0.90)
+INITIAL_GARCH_PARAMETERS = 0.05, 0.90
 
 # Copulas
 VINE_COPULA_NUMBER_OF_THREADS = 6
@@ -24,7 +24,7 @@
 
 # Extreme Value Theory POT Threshold
 EVT_THRESHOLD = 0.95
-GEV_INITIAL_SOLUTION = (0.1, 0.01)
+GEV_INITIAL_SOLUTION = 0.1, 0.01
 
 # Optimizer
 DEFAULT_SOLVER = "ipopt"

src/abacus/models/ar.py

@@ -1,7 +1,6 @@
 # -*- coding: utf-8 -*-
 import torch
 import numpy as np
-import pandas as pd
 
 from torch.distributions import Normal
 

src/abacus/models/garch.py

@@ -1,7 +1,6 @@
 # -*- coding: utf-8 -*-
 import torch
 import numpy as np
-import pandas as pd
 
 from scipy.optimize import minimize
 from torch.distributions import Normal

src/abacus/models/model.py

@@ -22,7 +22,7 @@ def calibrate(self):
         """Calibrates the parameters of the model. A non-calibrated model does not
         have an AIC/BIC and simulations are not possible.
         """
-        ...
+        pass
 
     @abstractmethod
     def transform_to_uniform(self):
@@ -31,7 +31,7 @@ def transform_to_uniform(self):
 
         Required for copula calibration.
         """
-        ...
+        pass
 
     @abstractmethod
     def transform_to_true(self, uniform_sample: torch.Tensor) -> torch.Tensor:
@@ -44,21 +44,21 @@ def transform_to_true(self, uniform_sample: torch.Tensor) -> torch.Tensor:
         Returns:
             torch.Tensor: Sample of simulated process in chronological order.
         """
-        ...
+        pass
 
     @property
     @abstractmethod
     def _log_likelihood(self):
         """
         """
-        ...
+        pass
 
     @property
     @abstractmethod
     def _number_of_parameters(self):
         """
         """
-        ...
+        pass
 
     @property
     def aic(self) -> torch.Tensor:

src/abacus/optimizer/optimizer.py

@@ -18,14 +18,15 @@ class OptimizationModel(ABC):
 
     _model_specification: ClassVar[str]
 
-    def __init__(self, universe: Universe, portfolio: Portfolio, simulation_tensor: torch.Tensor, solver: str=DEFAULT_SOLVER):
+    def __init__(self, universe: Universe, portfolio: Portfolio, simulation_tensor: torch.Tensor, solver: str=DEFAULT_SOLVER, verbose: bool=True):
         self._portfolio = portfolio
         self._simulation_tensor = simulation_tensor
         self._universe = universe
         self._asset_identifiers = universe.instrument_identifiers
         self._solver = solver
         self._solved = False
         self._ampl = None
+        self._verbose = verbose
 
     def solve(self):
         # TODO: Consider a verbose mode to display command line output.
@@ -46,7 +47,7 @@ def _initiate_ampl_engine(self):
 
     def _solve_optimzation_problem(self):
         self._check_initialization()
-        self._ampl.solve()
+        self._ampl.solve(verbose=self._verbose)
 
     def _check_solved(self):
         if not self._solved:

src/abacus/simulator/model_selector.py

@@ -13,7 +13,6 @@
 
 
 class ModelSelector:
-    # TODO: Add a selection strategy LOOCV, other information criterion, etc.
     # TODO: Add more models here.
 
     STOCK_ADMISSIBLE_MODELS = AR, GARCH

src/abacus/simulator/simulator.py

@@ -31,6 +31,7 @@ def covariance_matrix(self, data_type: DataTypes=DataTypes.LOG_RETURNS) -> torch
         # TODO: Add input for price through enum.
         # TODO: Consider making static.
         # TODO: Consider check for symmetric and positive semi definiteness.
+        # TODO: Consider having in universe? Does not actually pertain to simulations itself.
         if data_type == DataTypes.LOG_RETURNS:
             instrument_data = [instrument.log_returns_tensor for instrument in self._instruments]
         elif data_type == DataTypes.ART_RETURNS: