Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
9 Cards in this Set
- Front
- Back
|
Heteroscedasticity |
Variance of current errors being dependent on lagged variance
Unreliable Fstats Understated SE, inflated t stat It does not affect consistency of parameters but leads to mistakes in inference (that is conducting statistical tests)
If R^2 is close to 0, then prob no heteroscedasticity |
|
|
Consistency of parameters |
As n increases, it converges to true value |
|
|
Unconditional vs conditional |
Unconditional - error terms are not correlated with X term. Not a biggie.
Conditional - Error terms correlated with one or more X's. Biggie. -- Testing: Breusch pagan, White test -- correcting: Generalized least Squares, Robust standard error. |
|
|
Serial correlation |
DW = 0 means +ve serial correlation Dw= 2, no correlation DW = 4 -ve correlation
Error terms are correlated. High correlation between previous and next error terms.
Assumptions: That it takes First order form of serial correlation. This means, signs (-ve or +ve) of error term persists over time.
Consequences: 1) It does not affect consistency if parameter but affects infernece which is ability to conduct statistical test. 2) F-test is inflated because error MSE (SSE/n-k-1) is underestimated. 3) inflates t stats, underestimated standard error. Type I error.
Testing: 1) Adjust SE to account for serial correlation 2) Hansen, Newey and west method.
Note that DW stats can't be used in AR model that can't be used when Y is the lagged value of X's. |
|
|
What is covariance stationary? |
1)Mean (expected value of time series) 2)and variance 3)Covariance All do not change over time and are contact.
This assumptions need to hold true to conduct AR model. This means that lagged X's are random. Remember random walk is not Covariance stationary. |
|
|
When do we use RMSE? |
We compare out if sample forecasting performance by comparing root mean squares error. |
|
|
Imp diff between old and Ar model |
In ols we had to ensure that X's are not random. In At we have to ensure that x(t-1) which is X technically, are random. |
|
|
Probit and Logit model |
Used for models with qualitative dependent variables such as models with Y that can have two discrete outcomes (0 or 1).
Probit and Logit similar. Just the diff is probit is based on ND. Logit is based on "logistic" distribution model. |
|
|
Unit root |
Null hypothesis is that there is unit root. You want to reject null. Unit root means data is not staionary. Spurious results - high r^2 with uncorrelated data. First order diff is used to solve it. Dickey Fuller test is also used. Augmented dicky Fuller test is also used. |
