Weeks 7 & 8: Regression Flashcards
What does the X variable do?
X - Independent variable explains the changes in Y (dependent variable), by falling back on statistical power and validity
What is X variable?
Independent variable - does not change
What is Y variable?
Dependent variable - what we are trying to measure, influenced by X.
What kind of model is Regression?
Causal model - prediction on target time series is linked to other time series. You can causally explain why the prediction may be adequate
Advantages of Regression
- Can create scenario & strategy based predictions
- If assumptions satisfied & model correctly specified, OLS forecasts will be unbiased and efficient
- Can identify the relationship between two or more variables
Disadvantages of Regression
Requires more data, resources and theoretical knowledge
What happens with the within-sample-data?
Unlike traditional models (e.g MA), the within-data is NOT redundant. It will be extrapolated to make future forecasts
What are the OLS Assumptions?
- Correct functional form
CHECK RESIDUALS:
1. Zero mean
2. Constant variance
3. Independently derived
4. Randomly scattered
What is B0?
The intercept.
Estimated Yt when X = 0
What is B1?
The estimated constant increase in Yt when X increases by 1 unit.
What is R2?
Coefficient of determination is a measure of how well the model fits the data
-The proportion of variation in Y is accounted for by the X values
How to calculate R2?
SSR/SST
What is a good R2?
Closest to 1, but rule of thumb = 70%
What is Standard Error used for?
Used to compare model performance
How to calculate Standard Error?
SQRT(MSE)
What is Overall Model Significance?
F test (overall model fitness), shows if relationship between ALL X & Y variables exist
- If F Significance <0.05 - it has explanatory model power
How to test Individual Coefficients?
If the slope coefficient (B1) is 0, the X variable does not influence Y
- It could be an irrelevant variable and can be removed through Stepwise Regression
What is an Individual Variable Significance test?
T test - show if there’s a relationship between X and Y. If p-value below 0.05, variable has power to explain changes in y variable
The larger the t stat… the….?
lower the p value