# delimit ;
* Exercise 6: assessing the determinants of credit card concession policy;
use "C:\Users\AutoLogon\Desktop\creditcard.dta", clear;
log using "C:\Users\AutoLogon\Desktop\exercise6.log", replace;
* We will construct a binomial model to explain the characteristics of the applicants that determine the propability of having the demand accepted/refused. We specify a probit model.;
probit cardhldr majordrg age ownrent selfempl income depndt active cur_add;
* We exlude the two non-significant variables (age and cur_add) from our model and re-estimate it;
probit cardhldr majordrg ownrent selfempl income depndt active;
* We observe that applicants with previous records of payment default
(majordrg), which are self-employed and/or with a higher number of dependents have a lower probability of having the application accepted. On the other hand, people with higher income and/or which own their houses and/or have other credit cards active are associated to a higher probability of having the credit card accepted.
Remember that we can only interpret the sign of the coefficient in terms of marginal effects, but its value does not provide the magnitude of the change. To assess the magnitude of the marginal effect we need to use the mfx command;
mfx;
* An additional record of previous payment default decreases the probability of acceptance by 27 percentage points. Applicants that are self-employed have 13 percentage points lower probability of having the application accepted when compared to other applicants. Each additional (family) dependent decreases the porbability of having the application accepted by 3.9 percentage points. On the other hand, a 1,000 USD increase in yearly income is associated with a 3.1 percentage points increase in the probability of having the application accepted (similar
analysis to the other positive marginal effects);
* Constructing the hit rate table.;
estat class ;
* We observe that our model have correctly predicted 84.91% of the application decisions (=(1,000+120)/1319). LetÂ´s now explore some conditional probabilities.
Probab(model predicts refusal, given that application was effectively accepted) = Prob(-|D) = 23/1,023 = 2.25%
Probab (application was effectively refused given that model predicts to refuse) =Prob(~D|-) = 120/142 = 83.92%;
log close ;