基于提供的《Problem Set 1.docx》的内容，我将为每个问题概述Python代码的关键步骤和逻辑框架。请注意，实际的实现还需要依赖于具体的数据处理细节以及您选择使用的特定库版本等。以下是针对每个任务的大致代码结构： ### 1. (Naïve) Factor Regression 此部分需要进行滚动窗口回归分析，并绘制市场因子（`Mkt-RF`）及系数的时间序列图。对于PERMNO=10324的股票特别关注其结果解释。 ```python import pandas as pd from statsmodels.regression.linear_model import RollingOLS import matplotlib.pyplot as plt # Load data stock_returns = pd.read_csv('monthly_stock_returns.csv') ff_factors = ... # Load Fama-French 5 Factors hxz_factors = ... # Load Hou, Xue, Zhang q-Factors ps_factor = ... # Load Pastor Stambaugh Liquidity Factor # Merge datasets based on date merged_data = ... # Define the rolling window size window_size = 60 # Perform rolling OLS for PERMNO 10324 def perform_rolling_regression(stock_id): stock_data = merged_data[merged_data['PERMNO'] == stock_id] # Prepare X (factors) and y (returns) X = ... y = ... # Fit rolling OLS roll_reg = RollingOLS(y, X, window=window_size) rolling_results = roll_reg.fit() # Extract beta coefficients and intercepts betas = rolling_results.params return betas betas_per_stock = perform_rolling_regression(10324) # Plotting plt.figure(figsize=(14, 7)) plt.plot(betas_per_stock.index, betas_per_stock['Mkt-RF'], label='Mkt-RF Beta') plt.title("Time Series of Mkt-RF Beta for PERMNO 10324") plt.legend() plt.show() # Explain findings... ``` ### 2. Fama-MacBeth Regression 在此任务中，需执行Fama-MacBeth两步法以估计随时间变化的风险溢价，并绘制市场风险溢价(`Mkt-RF`)的变化趋势图。 ```python from linearmodels.panel import PanelOLS # Assuming `panel_data` contains the prepared panel data for Fama-MacBeth def fama_macbeth_regression(panel_data): # First step: Cross-sectional regression for each time period # Second step: Time-series average of the coefficients # Note: This simplified implementation assumes preprocessed data suitable for FM regression model = PanelOLS.from_formula('dependent_variable ~ independent_variables', data=panel_data) fm_results = model.fit(cov_type='clustered', cluster_entity=True) # Extract and analyze the risk premia over time risk_premiums = fm_results.params return risk_premiums risk_premiums = fama_macbeth_regression(panel_data) # Plotting the Mkt-RF risk premium over time plt.figure(figsize=(14, 7)) plt.plot(risk_premiums.index, risk_premiums['Mkt-RF'], label='Mkt-RF Risk Premium') plt.title("Time Series of Mkt-RF Risk Premium") plt.legend() plt.show() # Discuss findings... ``` ### 3. LASSO Regression 该部分采用LASSO方法代替普通最小二乘法来进行因子选择，并统计每个月出现频率最高的前五个因子。 ```python from sklearn.linear_model import Lasso def lasso_factor_selection(data, alpha=0.001): # Initialize storage for top factors per month top_factors_by_month = [] for year_month in data.groupby(['year', 'month']): X, y = prepare_X_y(year_month) # Function to prepare features and target # Apply LASSO lasso = Lasso(alpha=alpha) lasso.fit(X, y) # Identify non-zero coefficients (selected factors) selected_factors = [factor for factor, coef in zip(X.columns, lasso.coef_) if abs(coef) > 1e-6] # Collect top 5 most frequent factors freq_dist = pd.Series(selected_factors).value_counts().head(5) top_factors_by_month.append(freq_dist.index.tolist()) return pd.DataFrame(top_factors_by_month, columns=['1st', '2nd', '3rd', '4th', '5th']) lasso_results = lasso_factor_selection(prepared_data) print(lasso_results) # Experiment with different alphas... ``` 上述代码片段展示了如何使用Python来完成这三个主要的任务。当然，在实际操作时，你需要根据自己的数据集调整具体的函数定义和其他参数设置。此外，还需确保所有外部数据源正确加载并预处理成适合这些算法的形式。