L25 - Distributed Lags Flashcards

1
Q

What does a scatterplot look like for a binary dependent variable?

A
  • e.g. say if Yi = {0,1}
  • Then the points can only lie on the x-axis (y=0) or the line y=1 –> this makes it very hard to fit a linear equation to the scatterplot
  • So instead of fitting a linear line, we fit a logistic curve with a sigmoid shape
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

How can you interpret the coefficients in limited dependent variable models?

A
  • What is the expected value of Y|X?
    • Well if Y can only be 0 or 1 (binary) then its expected value must be ) and 1 multiplied by the probability of occurring ( which will just leave whatever is multiplied by 1)
    • The F(X) is the conditional probability of getting Y=1
    • The first derivative gives the marginal effect of a change in X on the expected value of Y –> NOT CONSTANT
  • F’(X) = β * Pr(Y=1) * Pr(Y=0)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

At what point do we evaluate a limited dependent variable model?

A
  • at the mean of the data (use X(bar) in the calculation for the marginal effect)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Can other functions be used to solve Limited Dependant variable models?

A
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What are Distributed Lags?

A

Distributed lags are dynamic relationships in which the effects of changes in some variable X on some other variable Y are spread through time.

Distributed lags can arise for a variety of reasons including:

  1. Costs of adjustment –> costly to change Y variable all at once
  2. The effects of expectations –> take a while for expectations to adjust

Below depicts two ways we can account for them in a regression

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

How do econometrician account for correlation between lagged X variable when using a Infinite distributed lag model?

A
  • With geometrically declining weights, the future we go into the past the less of an impact its weight will have thus beta for each coefficient is halved each time
    • It could be said that βi = β0λi and as long as λ < 1 (so it converges), will tend to zero –> the weights get less and less the further in the pas the lag is
    • this reduces the number of parameters we need to estimate from an infinite number of β’s to just two: β0 and λ
  • both have the same marginal effect –> the geometric sum of all the β in the infinite model = 1
    *
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

How can you estimate a geometrically declining weifghts model?

A
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

What is the adaptive expectation model?

A
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

What is the problem with the adaptive expectations model?

A
  • it is persistently and predictably wrong
    • it is systematic below actual inflation
How well did you know this?
1
Not at all
2
3
4
5
Perfectly