Hi Derek,
first of all, let me say I appreciate your work a lot, your Youtube tutorials are really helping me.
After watching your video "Stock Market Prediction: Python for Finance 5", I've been working with the code, and I think I found a serious problem. In section Regression Time Series Example with AutoReg, the regression model is initialized as
train_model = AutoReg(a_df['Adj Close'], 500).fit(cov_type="HC0")
The problem is that it uses a_df to train the model, the whole dataset instead of just train_df, the training dataset. This is a serious problem because this way we are also using test data to train the model, this is "cheating"!
I tried fixing this with train_model = AutoReg(train_df['Adj Close'], 500).fit(cov_type="HC0"), and played with the lag parameter, but the correctly trained model is a really bad predictor for test_df.
Hi Derek,
first of all, let me say I appreciate your work a lot, your Youtube tutorials are really helping me.
After watching your video "Stock Market Prediction: Python for Finance 5", I've been working with the code, and I think I found a serious problem. In section Regression Time Series Example with AutoReg, the regression model is initialized as
train_model = AutoReg(a_df['Adj Close'], 500).fit(cov_type="HC0")The problem is that it uses
a_dfto train the model, the whole dataset instead of justtrain_df, the training dataset. This is a serious problem because this way we are also using test data to train the model, this is "cheating"!I tried fixing this with
train_model = AutoReg(train_df['Adj Close'], 500).fit(cov_type="HC0"), and played with thelagparameter, but the correctly trained model is a really bad predictor fortest_df.