带最优滞后阶数的VAR多变量预测：72列数据批量处理需求

Got it, let's tackle this problem step by step. Since you need to handle 72 endogenous variables (like dj1 to dj72) and generate point predictions for each, here's a practical, scalable approach tailored to your needs:

Step 1: Automate Endogenous Variable List #

First, skip manually typing 72 variable names—generate the list programmatically:

# Create a list of your 72 endogenous variable names (dj1 to dj72)
endog_vars = [f"dj{i}" for i in range(1, 73)]

Step 2: Initialize a Results Container #

We'll use a pandas DataFrame to store all predictions neatly:

import pandas as pd
import statsmodels.api as sm  # Adjust based on your preferred model library

Step 3: Loop Through Each Endogenous Variable #

For each dj variable, fit your model, generate point predictions, and store the results. Below is a generic example using OLS (swap in your actual model type—ARIMA, GLM, etc.—as needed):

# Initialize empty DataFrame to hold predictions (matches your original data's index)
predictions_df = pd.DataFrame(index=dataframe.index)

for var in endog_vars:
    # Define your model's endogenous (y) and exogenous (X) variables
    y = dataframe[var]
    # Example: Use all non-dj columns as exogenous data (adjust to your actual equation spec!)
    X = dataframe.drop(columns=endog_vars)
    X = sm.add_constant(X)  # Add intercept if your model requires it
    
    # Fit the model
    model = sm.OLS(y, X).fit()
    
    # Generate point predictions (in-sample shown; use out-of-sample data if needed)
    var_predictions = model.predict(X)
    
    # Add predictions to our results DataFrame
    predictions_df[var] = var_predictions

Step 4: Get Predictions as Matrix (If Needed) #

If you need a matrix instead of a DataFrame, just convert it:

predictions_matrix = predictions_df.values

Bonus: For Multi-Variable Models (Like VAR) #

If your 72 dj variables are modeled together (e.g., Vector Autoregression), skip the loop and fit a single multi-variable model for efficiency:

from statsmodels.tsa.vector_ar.var_model import VAR

# Subset your data to only the endogenous variables
endog_data = dataframe[endog_vars]

# Fit the VAR model (adjust maxlags to your needs)
var_model = VAR(endog_data)
fitted_var = var_model.fit(maxlags=2)

# Generate point predictions (e.g., 1 step ahead; adjust steps as needed)
var_predictions = fitted_var.forecast(fitted_var.y, steps=1)

# Convert to DataFrame for readability
predictions_df = pd.DataFrame(var_predictions, columns=endog_vars)

Key Notes #

• Swap out the model fitting code (sm.OLS, VAR, etc.) with whatever model you're actually using for your equations.
• Adjust the exogenous variable selection (X = dataframe.drop(...)) to match your specific equation specifications.
• For out-of-sample predictions, pass your out-of-sample exogenous data to the predict() method instead of the in-sample X.

内容的提问来源于stack exchange，提问作者DaWassi