R help - Apr 2013 - Iterative regression through a series

Hello, 

Some context:  let's say I have a data series (let's call it PRICE, for
simplicity), sample size N.  I have a desire to regress that on TIME, and then
to use the TIME and intercept coefficients to predict the price in the next
period and to use the standard error to calculate a confidence interval.  This
is all very easy.

However, what I need help for is to calculate a confidence interval for each
point in time:  imagining that at the end of the 10th period I have 10 data
points, and wish to regress them on the 10 periods to create a confidence
interval for the next 'predicted' price.  And so on from TIME[10:100].
 So the first regression would be of PRICE[1:10] on TIME[1:10], the second of
PRICE[1:11] on TIME[1:11], the third of PRICE[1:11] on TIME[1:11], and so on to
PRICE[1:N] and TIME[1:N].  I'd like to be able to vary the starting point
(so it would need to be an argument in the function, in this case it would be
10).  The ultimate output of the code would be to save the TIME coefficients and
standard errors it generates to two vectors, say TIME.coef and TIME.SE.

I'm not sure if lapply() can be bent to my will, or if a for loop would be
too inefficient, or what.  I'm not new to R, but I'm fairly new to this
kind of programming.

This is a bungled mess of a narrative, and I apologize.  Please feel free to use
TIME=1:100 and PRICE=rnorm(100,8,2).

Here's an attempt, which has failed for reasons I can only imagine.  Any
help getting this to work would be greatly appreciated.  Any help doing this
without loops would be even better.

> Time=1:100 
> Price=rnorm(100,8,2) 
> 
> estim=0     #I'm hoping this will be the Time coefficient 
> error=0      #I'm hoping this will be the standard error of the Time
coefficient
> for (i in Time) { 
+     regr=lm(Price[1:i]~Time[1:i]) 
+     estim=c(estim,coef(summary(regr))[2,1]) 
+     error=c(error,coef(summary(regr))[2,1]) 
+     } 
Error: subscript out of bounds


Many, many thanks in advance. 

Mendi 
	[[alternative HTML version deleted]]

Hello,

The error comes from NAs where you would expect coefficients. Try the 
following.



set.seed(7511)  # Make the example reproducible

N <- 100
Time <-1:N
Price <- rnorm(N, 8, 2)

estim <- numeric(N)   # It's better to allocate the results
error <- numeric(N)   # vectors in advance
for (i in Time) {
      regr <- lm(Price[1:i] ~ Time[1:i])
      estim[i] <- coef(regr)[2]
      if(is.na(coef(regr)[2]))
	     error[i] <- NA
      else
	     error[i] <- summary(regr)$coefficients[2,2]

}


Hope this helps,

Rui Barradas

Em 02-04-2013 17:44, triskell4-umbrella at yahoo.fr
escreveu:
> Hello,
>
> Some context:  let's say I have a data series (let's call it PRICE,
for simplicity), sample size N.  I have a desire to regress that on TIME, and
then to use the TIME and intercept coefficients to predict the price in the next
period and to use the standard error to calculate a confidence interval.  This
is all very easy.
>
> However, what I need help for is to calculate a confidence interval for
each point in time:  imagining that at the end of the 10th period I have 10 data
points, and wish to regress them on the 10 periods to create a confidence
interval for the next 'predicted' price.  And so on from TIME[10:100]. 
So the first regression would be of PRICE[1:10] on TIME[1:10], the second of
PRICE[1:11] on TIME[1:11], the third of PRICE[1:11] on TIME[1:11], and so on to
PRICE[1:N] and TIME[1:N].  I'd like to be able to vary the starting point
(so it would need to be an argument in the function, in this case it would be
10).  The ultimate output of the code would be to save the TIME coefficients and
standard errors it generates to two vectors, say TIME.coef and TIME.SE.
>
> I'm not sure if lapply() can be bent to my will, or if a for loop would
be too inefficient, or what.  I'm not new to R, but I'm fairly new to
this kind of programming.
>
> This is a bungled mess of a narrative, and I apologize.  Please feel free
to use TIME=1:100 and PRICE=rnorm(100,8,2).
>
> Here's an attempt, which has failed for reasons I can only imagine. 
Any help getting this to work would be greatly appreciated.  Any help doing this
without loops would be even better.
>
>
>> Time=1:100
>> Price=rnorm(100,8,2)
>>
>> estim=0     #I'm hoping this will be the Time coefficient
>> error=0      #I'm hoping this will be the standard error of the
Time coefficient
>> for (i in Time) {
> +     regr=lm(Price[1:i]~Time[1:i])
> +     estim=c(estim,coef(summary(regr))[2,1])
> +     error=c(error,coef(summary(regr))[2,1])
> +     }
> Error: subscript out of bounds
>
>
> Many, many thanks in advance.
>
> Mendi
> 	[[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.
>