Add files via upload

Evolutionary Portfolio Optimisation/cma_exp.py

@@ -0,0 +1,138 @@
+#%% Import libraries
+import numpy as np
+import cma
+import yfinance as yf
+from datetime import datetime
+import warnings
+from matplotlib import pyplot as plt
+
+np.set_printoptions(suppress=True)
+warnings.simplefilter(action='ignore', category=FutureWarning)
+warnings.simplefilter(action='ignore', category=UserWarning)
+
+#% Define functions
+def calc_returns(prices):
+    return np.log(prices/prices.shift(1)).dropna()
+
+def expected_returns(weights, mean_returns):
+    return np.dot(weights, mean_returns)
+
+def portfolio_std(weights, returns_covmat):
+    return np.sqrt(np.dot(weights.T, np.dot(returns_covmat, weights)))
+
+def sharpe_ratio(weights, mean_returns, returns_covmat):
+    weights /= np.sum(weights)
+    R = expected_returns(weights, mean_returns)
+    sigma_p = portfolio_std(weights, returns_covmat)
+    return (R) / sigma_p
+
+def objective_function(weights, mean_returns, returns_covat, alpha_penalty):
+    fitness = sharpe_ratio(weights, mean_returns, returns_covat) 
+    fitness += alpha_penalty * min(0, weights.min())
+    return -1*fitness
+
+#% Import data
+start_date = datetime(2009, 3, 1)
+end_date = datetime(2024,3,1)
+
+# stocks = 'AAPL AMZN GLD GOOGL GS META NFLX PFE SIEGY TM'
+stocks = '''AAPL ABT ADBE AMD AMZN AXP BA BABA BAC CSCO CVS DHR DIS F GLD GOOGL 
+            GS HD HON IBM INTC JNJ KO LLY LVMUY MA MCD META MMM MSFT NFLX NKE 
+            NVDA ORCL PFE PG PYPL QCOM RIO RTX SAP SBUX SIEGY T TM TMO TSLA 
+            UNH V WMT'''
+
+prices = yf.Tickers(stocks).history(start = start_date, end = end_date)['Close']
+
+daily_returns = calc_returns(prices)
+pre_agg_returns = daily_returns.reset_index().drop(columns='Date')
+weekly_returns = pre_agg_returns.groupby(pre_agg_returns.index // 5).sum()
+monthly_returns = pre_agg_returns.groupby(pre_agg_returns.index // 22).sum()
+
+returns = daily_returns
+mean_returns = returns.mean()
+returns_covmat = returns.cov()
+
+#%% Optimisation
+# CMA-ES
+num_assets = returns.shape[1]
+initial_weights = np.ones(num_assets) / num_assets  
+initial_sigma = 1
+alpha_penalty = 100
+
+es = cma.CMAEvolutionStrategy(initial_weights, initial_sigma)
+
+mean_history = []
+
+while not es.stop():
+    candidate_solutions = es.ask()
+    objective_values = [
+        objective_function(s, mean_returns, returns_covmat, alpha_penalty)
+        for s in candidate_solutions]
+    es.tell(candidate_solutions, objective_values)
+    es.disp()
+
+    mean_history.append(es.mean)
+
+cma_output = np.round(es.result.xbest, 5)
+cma_solution = list(zip(stocks.split(), cma_output))
+cma_sharpe = -es.result.fbest
+
+# Portfolio Opt. (Benchmark)
+from pypfopt.efficient_frontier import EfficientFrontier
+ef = EfficientFrontier(mean_returns, returns_covmat)
+weights = ef.max_sharpe(risk_free_rate = 0.0)
+benchmark_solution = list(ef.clean_weights().items())
+benchmark_sharpe = ef.portfolio_performance(risk_free_rate = 0.0)[2]
+
+#%% Results
+# Plotting the parameter history of top v parameters. 
+# v = 2
+# top_index = np.argsort(cma_output)[-v:][::-1]
+# mean_history = np.array(mean_history)
+# colors = ('r', 'b', 'g', 'o')
+
+# plt.figure(figsize=(10, 5))
+# for i in range(len(top_index)):
+#     j = top_index[i]
+#     stock = np.array(stocks.split(), dtype="str")[j].item()
+#     plt.plot(mean_history[:, j], color = colors[i], label=f'Weight for {stock}')
+#     plt.axhline(y=cma_output[j], color = colors[i] ,linestyle = '--', label=f'Best weight for {stock}')
+
+# plt.xlabel('Generation')
+# plt.ylabel('Weight value')
+# plt.title(f'Evolution of Top {v} Weights')
+# plt.legend()
+# plt.show()
+
+print("CMA-ES Solution:", cma_solution)
+print("CMA-ES Sharpe Ratio:", cma_sharpe)
+print("Benchmark Solution:", benchmark_solution)
+print("Benchmark Sharpe Ratio:", benchmark_sharpe)
+
+data = [cma_solution, benchmark_solution]
+tickers = [d[0] for d in data[0]]  # Get the tickers from the first data entry
+values1 = [d[1] for d in data[0]]  # Values from the first dataset
+values2 = [d[1] for d in data[1]]  # Values from the second dataset
+
+# Create a new figure
+fig, ax = plt.subplots(figsize=(10, 4))
+
+# Number of bars
+n = len(tickers)
+index = range(n)
+
+bar_width = 0.35
+
+bars1 = ax.bar(index, values1, bar_width, label='CMA-ES Weights', color = 'r')
+bars2 = ax.bar([p + bar_width for p in index], values2, bar_width, label='Benchmark Weights', color = 'b')
+
+ax.set_xlabel('Stock Ticker')
+ax.set_ylabel('Weight')
+ax.set_title('CMA-ES vs Benchmark – Medium Term Strategy')
+ax.set_xticks([p + bar_width / 2 for p in index])
+ax.set_xticklabels(tickers)
+ax.legend()
+
+plt.show()
+
+# %%