代写problem set、代做R设计、R编程语言调试、代写datasets

日期：2019-10-23 11:15

Problem Set 3

In this problem set, you will estimate regression models with a binary dependent variable

with the goal of predicting whether American Airlines enters a given market.

You are allowed to work in groups of up to three students, but you must disclose the

members of your group. Individual submissions are required. The code you submit may be

identical to the one of the other group members, but we expect the comments and answers

to the questions to be your own.

Your submission should consist of: (i) a R markdown document with code, figures and

comments, and (ii) a zip folder containing all files needed for replication. You may upload

these materials via the course’s Canvas website (please do not email us with your homework

submission).

Each of the four questions below summarize the required tasks, with the individual

bullet points detailing the steps needed to complete them.

0) Download the market-level and market-airline-level datasets,

"market_level.R"

and

"market_airline_level.R".

Set the seed to 0 and randomly allocate 1,000 rows of the market-level data to a test

set, to be used only in (7). Use the rest to do the following.

(1) Estimate a linear probability model, predicting whether American Airlines enters a

market as a function of the number of competitors. Note: American Airlines’ ticket carrier

id is “AA”.

2) Repeat (1) using a logit model instead of a linear probability model.

1

3) Repeat (1) using a probit model instead of a linear probability model.

4) Compute non-parametric estimates of the conditional probabilities of entering. (ie

compute the conditional probability of entering conditional on each number of competitors

directly from the data).

5) Plot the fitted values of each regression in one graph (i.e. estimated probabilities

on the y-axis and the number of competitors on the x-axis). In words, explain the coeffi-

cients of the first three models. How do the estimated relationships compare? Should we

interpret these relationships causally? Are the estimates for the probit and logit similar?

Should we have expected this ex ante?

6) Obviously other covariates matter in predicting whether or not American will enter

a particular route. In addition to the number of competitors, add the average market

distance, market size, hub route indicator, vacation route indicator, slot controlled indicator,

and market income to the set of predictors. Fit to the data L1 regularized logistic

regression (ie Lasso for logit, pg 125-126 ESL) where the full model includes all squared

terms and second-order cross terms. Using a 10-fold cross validation procedure, find the

optimal value of lambda.

7) Calculate the MSE on the test set for each of your 5 models and put them in a table.

Explain your results.

2