Multiple regressions with changing dependent variable and time span

Hi,
#####1 & 2:

set.seed(432)
dat1 <- as.data.frame(matrix(sample(c(1:60,NA),154*337,replace=TRUE),ncol=337))
?colnames(dat1) <- c(paste("F",1:3,sep="."),paste("r",1:334,sep="."))
lst1 <- lapply(paste("r",1:334,sep="."),function(x) cbind(dat1[,c(1:3)],dat1[x]))

?lst2 <- lapply(lst1,function(x) {colnames(x)[4] <-"r";x} )
?
?sapply(lst2,function(x) sum(!!rowSums(is.na(x))))
library(zoo)
rollapply(lst2[[1]],width=32,FUN=function(z) {z1 <- as.data.frame(z); sum(!!rowSums(is.na(z1)))},by.column=FALSE,align="right")
?#[1] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 3 3 3 3
?#[38] 4 4 4 4 4 4 4 4 4 3 3 3 3 4 4 4 4 4 4 4 5 5 5 4 4 4 4 4 3 3 3 3 2 2 2 2 2
?#[75] 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 0 0 0 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2
#[112] 2 2 2 2 2 2 2 2 2 2 2 2

?
res1 <- do.call(rbind,lapply(lst2,function(x) rollapply(x,width=32,FUN=function(z) {z1 <- as.data.frame(z); if(!sum(!!rowSums(is.na(z1)))) {l1 <-lm(r~F.1+F.2+F.3,data=z1); c(coef(l1), pval=summary(l1)$coef[,4], rsquare=summary(l1)$r.squared) } else rep(NA,9)},by.column=FALSE,align="right")))
row.names(res1) <- rep(paste("r",1:334,sep="."),each=123)
?dim(res1)
#[1] 41082???? 9


###3. 


library(car)

# vif()
res2 <- do.call(rbind,lapply(lst2,function(x) rollapply(x,width=32,FUN=function(z) {z1 <- as.data.frame(z); if(!sum(!!rowSums(is.na(z1)))) {l1 <-lm(r~F.1+F.2+F.3,data=z1); vif(l1) } else rep(NA,3)},by.column=FALSE,align="right")))
row.names(res2) <- rep(paste("r",1:334,sep="."),each=123)
dim(res2)
#[1] 41082???? 3


#DW statistic:
?lst3 <- lapply(lst2,function(x) rollapply(x,width=32,FUN=function(z) {z1 <- as.data.frame(z); if(!sum(!!rowSums(is.na(z1)))) {l1 <-lm(r~F.1+F.2+F.3,data=z1); durbinWatsonTest(l1) } else rep(NA,4)},by.column=FALSE,align="right"))
?res3 <- do.call(rbind,lapply(lst3,function(x) x[,-4]))
row.names(res3) <- rep(paste("r",1:334,sep="."),each=123)
?dim(res3)
#[1] 41082???? 3

##ncvTest()
f4 <- function(meanmod, dta, varmod) {
assign(".dta", dta, envir=.GlobalEnv)
assign(".meanmod", meanmod, envir=.GlobalEnv)
m1 <- lm(.meanmod, .dta)
ans <- ncvTest(m1, varmod)
remove(".dta", envir=.GlobalEnv)
remove(".meanmod", envir=.GlobalEnv)
ans
}

?lst4 <- lapply(lst2,function(x) rollapply(x,width=32,FUN=function(z) {z1 <- as.data.frame(z); if(!sum(!!rowSums(is.na(z1)))) {l1 <-f4(r~.,z1) } else NA},by.column=FALSE,align="right"))
names(lst4) <- paste("r",1:334,sep=".") 
length(lst4)
#[1] 334


###jarque.bera.test
library(tseries)
res5 <- do.call(rbind,lapply(lst2,function(x) rollapply(x,width=32,FUN=function(z) {z1 <- as.data.frame(z); if(!sum(!!rowSums(is.na(z1)))) {l1 <-lm(r~F.1+F.2+F.3,data=z1); resid <- residuals(l1); unlist(jarque.bera.test(resid)[1:3]) } else rep(NA,3)},by.column=FALSE,align="right")))
?dim(res5)
#[1] 41082???? 3


##the lag() thing is not clear.

A.K.

Hi,

Thanks for reply!


Three things:
1. 
I did not write that some of the data has more then 31 NA in the column and then it is not possible to run lm()

Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) :??0 (non-NA) casesIn this case program should return "NA" symbol and go further, in the case when length of the observations is shorter then 31 program should always return "NA" but go further .


2. in your result matrix there are only 4 columns (for estimates of the coefficients), is it possible to put there 4 more columns with p-values and one column with R squared


3. basic statistical test for the regressions:

inflation factors can be captured by:
res2 <- do.call(rbind,lapply(lst2,function(x) rollapply(x,width=32,FUN=function(z) 
? vif(lm(r~ F.1+F.2+F.3,data=as.data.frame(z))),by.column=FALSE,align="right")))

and DW statistic:
res3 <- do.call(rbind,lapply(lst2,function(x) rollapply(x,width=32,FUN=function(z) 
? durbinWatsonTest(lm(r~ F.1+F.2+F.3,data=as.data.frame(z))),by.column=FALSE,align="right")))


3a)is that right? 
3b) how to do and have in user-friendly form durbinWatsonTest for more then 1 lag?

3c) how to apply: jarque.bera.test from library(tseries) and ncvTest from library(car) ???






Pozdrowienia,

Tomasz Schabek

Hi,

The link seems to be not working.? From the description, it looks like:
set.seed(432)
dat1 <- as.data.frame(matrix(sample(200,154*337,replace=TRUE),ncol=337))
?colnames(dat1) <- c(paste("F",1:3,sep="."),paste("r",1:334,sep="."))
lst1 <- lapply(paste("r",1:334,sep="."),function(x) cbind(dat1[,c(1:3)],dat1[x]))

?lst2 <- lapply(lst1,function(x) {colnames(x)[4] <-"r";x} )
library(zoo)

res <- do.call(rbind,lapply(lst2,function(x) rollapply(x,width=32,FUN=function(z) coef(lm(r~ F.1+F.2+F.3,data=as.data.frame(z))),by.column=FALSE,align="right")))

row.names(res) <- rep(paste("r",1:334,sep="."),each=123)
?dim(res)
#[1] 41082???? 4

coef(lm(r.1~F.1+F.2+F.3,data=dat1[1:32,]) )
#(Intercept)???????? F.1???????? F.2???????? F.3
#109.9168150? -0.1705361? -0.1028231?? 0.2027911
coef(lm(r.1~F.1+F.2+F.3,data=dat1[2:33,]) )
#(Intercept)???????? F.1???????? F.2???????? F.3
#119.3718949? -0.1660709? -0.2059830?? 0.1338608
res[1:2,]
#??? (Intercept)??????? F.1??????? F.2?????? F.3
#r.1??? 109.9168 -0.1705361 -0.1028231 0.2027911
#r.1??? 119.3719 -0.1660709 -0.2059830 0.1338608

A.K.






On Friday, November 29, 2013 6:43 PM, nooldor <nooldor at gmail.com> wrote:
Hi all!


I am just starting my adventure with R, so excuse me naive questions.

My data look like that:

<http://r.789695.n4.nabble.com/file/n4681391/data_descr_img.jpg>

I have 3 independent variables (F.1, F.2 and F.3) and 334 other variables
(r.1, r.2, ... r.334) - each one of these will be dependent variable in my
regression.

Total span of the time is 154 observations. But I would like to have rolling
window regression with length of 31 observations.

I would like to run script like that:

summary(lm(r.1~F.1+F.2+F.3, data=data))
vif(lm(r.1~F.1+F.2+F.3, data=data))

But for each of 334 (r.1 to r.334) dependent variables separately and with
rolling-window of the length 31obs.

Id est:
summary(lm(r.1~F.1+F.2+F.3, data=data)) would be run 123 (154 total obs -
31. for the first regression) times for rolling-fixed period of 31 obs.

The next regression would be:
summary(lm(r.2~F.1+F.2+F.3, data=data)) also 123 times ... and so on till
summary(lm(r.334~F.1+F.2+F.3, data=data))

It means it would be 123 x 334 regressions (=41082 regressions)

I would like to save results (summary + vif test) of all those 41082
regressions in one read-user-friendly file like this given by e.g command
capture.output()

Could you help with it?

Regards,

T.S.

??? [[alternative HTML version deleted]]

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.