OECD Economic Outlook

More data on current economy and projection for 2010
OECD Economic Outlook

For those applicants

For those applicants that still wait desperately for any kind of admission, may be the following article can relax you a little bit.
Even in such a competitive year, wish everyone gets his best luck!!
waiting list

Reputation

Paul Krugman
How to regain American reputation

Ranking of Economics Graduate School

USnews

1	Massachusetts Institute of Technology Cambridge, MA			Distance Enter your zip	Score 5.0
1	University of Chicago Chicago, IL			Distance Enter your zip	Score 5.0
3	Harvard University Cambridge, MA			Distance Enter your zip	Score 4.9
3	Princeton University Princeton, NJ			Distance Enter your zip	Score 4.9
3	Stanford University Stanford, CA			Distance Enter your zip	Score 4.9
3	University of California--Berkeley Berkeley, CA			Distance Enter your zip	Score 4.9
7	Yale University New Haven, CT			Distance Enter your zip	Score 4.8
8	Northwestern University Evanston, IL			Distance Enter your zip	Score 4.6
9	University of Pennsylvania Philadelphia, PA			Distance Enter your zip	Score 4.5
10	University of California--San Diego La Jolla, CA			Distance Enter your zip	Score 4.3

What's next?

Housing, Auto, Creditcard, then what?

Price control

"In many cases rent control appears to be the most efficient technique presently known to destroy a city -- except for bombing," said Swedish economist Assar Lindbeck, a former socialist who has been researching rent control since the 1960s.

Lessons from Great Depression

From CEA chair

Insight on current economy

Short run or long run?

Terriable time to reject skilled workers

More restriction on H1-B workers

Logistic regression and tree model

##read and clean the input file (deleting obs with missing values)
hmeq<-read.csv("hmeq.csv",na.strings="",h=T)
dim(hmeq)
dc<-hmeq[complete.cases(hmeq),]
dim(dc)
names(dc)

##randomly split the data and carry out logistic regression and tree model
fit<-c()
valid.fit<-c()
pred<-c()
tree.fit<-c()
tree.valid<-c()
tree.pred<-c()
for (i in c (1:10)){
samp.t<-sample(1:3364,1682)
train<-dc[samp.t,]
dim(train)
rest<-dc[-samp.t,]
samp.v<-sample(1:1682,841)
valid<-rest[samp.v,]
dim(valid)
test<-rest[-samp.v,]
dim(test)
##logistic regression model with binary response
lreg<-glm(BAD~.,data=train,binomial)
summary(lreg)
fit<-cbind(fit,lreg$fit)
valid.fit<-cbind(valid.fit,predict(lreg,valid))
pred<-cbind(pred,predict(lreg,test))
##recall"tree"command from "tree package"
library(tree)
hmeq.tree<-tree(BAD~.,data=train)
summary(hmeq.tree)
hmeq.tree
##plot the tree model
plot(hmeq.tree)
text(hmeq.tree,pretty=0,cex=0.8)
dev.off()
tree.fit<-cbind(tree.fit,hmeq.tree$y)
tree.valid<-cbind(tree.valid,predict(hmeq.tree,valid))
tree.pred<-cbind(tree.pred,predict(hmeq.tree,test))
}
##loop-calculate the train,validating,testing error from above two models with 10 replications
train_error<-c()
valid_error<-c()
test_error<-c()
f_error<-c()
v_error<-c()
t_error<-c()
for (i in 1:10){
fit1<-fit[,i]>0.5
fit.table<-table(fit1,train$BAD)
train_error<-c(train_error,(fit.table[1,2]+fit.table[2,1])/sum(fit.table))
train_error
valid.fit1<-valid.fit[,i]>0.5
valid.table<-table(valid.fit1,valid$BAD)
valid_error<-c(valid_error,(table[1,2]+table[2,1])/sum(table))
valid_error
pred1<-pred[,i]>0.5
table<-table(pred1,test$BAD)
test_error<-c(test_error,(table[1,2]+table[2,1])/sum(table))
test_error
table.f<-table(tree.fit[,i],train$BAD)
table.v<-table(tree.valid[,i],valid$BAD)
table.t<-table(tree.pred[,i],test$BAD)
f_error<-c(f_error,(table.f[1,2]+table.f[2,1])/sum(table.f))
v_error<-c(v_error,(table.v[1,2]+table.v[2,1])/sum(table.v))
t_error<-c(t_error,(table.t[1,2]+table.t[2,1])/sum(table.t))
}
##display the results errors from two models
l_errors<-cbind(train_error,valid_error,test_error)
t(l_errors)
t_errors<-cbind(f_error,v_error,t_error)
t(t_errors)

Best model of knn.cv method

##recall “knn.cv” from the package “class”

library(class)

##function-find best model of kNN based on v-fold cv

cv.k<-function(data,cl,k){

y<-(data)

knn.pred<-cv.err<-c()

for (i in 1:k){

knn.pred<-cbind(knn.pred,as.vector(knn.cv(y,cl,k=i)))

cv.err<-c(cv.err,sum(knn.pred[,i]!=class)/nrow(y))

}

opt.k<-which.min(cv.err)

print(opt.k)

}

opt.k<-cv.k(x,class,k=50)

opt.k

*****************************************************

##loop-find the best model of kNN method based on v-fold cv

knn.pred<-c()

cv.err<-c()

for (i in 1:50){

knn.pred<-cbind(knn.pred,as.vector(knn.cv(x,class,k=i)))

cv.err<-c(cv.err,sum(knn.pred[,i]!=class)/3364)

}

opt.k<-which.min(cv.err)

print(opt.k)

My first day here

I will try to learn something new every day, so this place will be a witness for my progress.
Goals--
Expertise in SAS, R, and Stata, at least one of them
Learn to analyze economy from economic pointview

Recently, I am struggling to write a function for finding best subsets models in R, and just at that time I found the following great place:
Experience with R--http://statisticsr.blogspot.com/
It has a lot of detailed notes for data mining programs.