Econometrics Notes
Dummy Variables
- Qualitative variables are introduced into a regression model as dummy variables.
- A dummy variable takes the value 1 if a qualitative attribute is present, and 0 if it is absent.
- Dummy variables classify a sample into categories based on attributes.
- The base category is the attribute assigned the value 0.
- If the regression has a constant term, the number of dummy variables must be one less than the number of qualitative variable categories.
- Dummy variables are used in analyzing structural stability and in regressions of pooled time series cross-sectional data.
Example of Dummy Variables
- Equation:
- Where:
- = grade of student i
- = 1 if i is male, 0 if i is female
- Where:
- - mean value of when i is female
- - mean value of when i is male
- = mean grade of female students
- = mean grade of male students, “slope “ is the difference between the mean grades of male and female students.
- t-test: H0: , indicating no significant difference between the mean grades of male and female students.
Dummy Variable Extended Example
- Equation:
- Where:
- = econometrics test score of student i
- = math ability of student i
- = sex of student i (1 = male, 0 = female)
- Where:
- Math ability is measured by the student’s grade in calculus.
- Hypothesis: higher calculus grade predicts higher econometrics test score, ceteris paribus; expected to be positive.
- The mean value of if i is female is
- The mean value of if i is male is
- Assumption: the relationship between the econometrics test scores and math ability is the same for male and female students.
- For a given math ability, the mean male and mean female econometrics test scores differ by
- t-test on (H0: ), to test if there is no significant difference in average econometrics scores between males and females.
Lag Models
- A lagged regression model is one where the dependent variable is affected by the lagged values of the regressor.
- A regression model in which the lagged values of the dependent variable are regressors is called an autoregressive model.
- When the dependent variable is explained only by the current and lagged values of an explanatory variable, the model is called a distributed lag model.
- If the dependent variable is a time series, its present value likely depends on its past values (autocorrelated), then it is logically to include lagged values of this dependent variable as explanatory variables and this is the main idea of time series models.
Details of Lag Models
- Equation:
- t = particular period in time
- An explanatory variable might affect the dependent variable after a certain length of time, lag.
- One-period lag is t-1, t-2 two period, current is t
- Autoregressive models predict future values based on past values. They are widely used in technical analysis to forecast future security prices. It implicitly assume that the future will resemble the past.
- A distributed lag model is a model that shows that investment reacts in a distributed way to changes in demand, not adjusting immediately.
Estimation Challenges
- A distributed lag model may be estimated by OLS by trimming it into a finite number of lags.
- However, the number of lags is an arbitrary choice and the more lags there are, the more degrees of freedom are lost. Moreover, the lagged values of the explanatory variable are likely to be highly correlated giving rise to multicollinearity.
- To avoid these problems, Koyck assumed that the coefficients of the lagged regressors decline geometrically, transforming the distributed lag model into an autoregressive model.
- Other auto regressive models have been formulated based on theory, two of which are the partial adjustment model and the adaptive expectations model.
Koyck, Partial Adjustment, and Adaptive Expectations Models
- All three models have a lagged dependent variable as regressors.
- In the Koyck and adaptive expectations models, the lagged dependent variable regressor and the error term are correlated resulting in biased and inconsistent OLS estimators.
- In the partial adjustment model, the OLS estimators are consistent.
ARDL Model
- An ARDL (Autoregressive-distributed lag) is a parsimonious infinite lag-distributed model.
- The term “autoregressive” shows that along with getting explained by the xt, yt also gets explained by its own lag also.
- Equation of ARDL(m,n) is as follows:
- Where:
- m and n are the numbers of years for lag
- is the disturbance terms
- ’s are coefficients for short-run relationship
- ’s are coefficients for long-run relationship.
- Where:
Granger Causality
- Lagged variables as regressors are useful in examining the Granger-Causality between variables.
- A variable X is said to Granger-cause a variable Y if X changes invariably before Y does.
- A Granger-causality test has been developed but care must be exercised in using the test since it is sensitive to the number of lags used in the test.
Stochastic Time Series
- A stochastic time series is one where the values of a variable are drawn from a probability distribution.
- Some of the important stochastic time series include the white noise, random walk, autoregressive models, and moving-average models.
- A stochastic time series is said to be stationary if the mean and variance are constant and the covariance between two terms of the series is a function only of the number of periods between them.
- The stationarity of a stochastic time series may be determined graphically by plotting the sample correlogram.
- A sample correlogram that declines gradually to zero indicates nonstationarity while one that drops quickly to zero indicates stationarity.
Stationarity
Stationarity may be tested formally by testing for the unit roots of the stochastic time series.
A stochastic time series that has a unit root is nonstationary.
The Dickey-Fuller Test is used to test for unit roots.
A series that has to be differenced of order d. A stationary series is I(0). Many economic time series are either I(0), I(1), or, occasionally, I(2).
Stationarity has implications to regression modelling. Many results in classical linear regression modelling are based on the assumption that the error term is a white noise, hence, stationary.
This means, for example, that in regression
the linear combination must be stationary. There is no problem if both and are I(0). But if they are I(I), then one has to check the stationarity of .
The stationarity of is tested by the Engle-Granger Test, a Dickey-Fuller test applied to the residuals of the above equation which is called the cointegrating equation.