Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: plot confidence intervals with time series


From   annoporci <annoporci@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: plot confidence intervals with time series
Date   Sat, 22 Dec 2012 23:11:19 +0800

On Sat, 22 Dec 2012 20:03:37 +0800, Christopher Baum <kit.baum@bc.edu> wrote:

Seems twoway lfitci is not too smart about format of the X-variable, so you need to tell it:
  graph twoway (lfitci $x1 $t, xlab(,format(%td)))  (scatter $x1 $t)



Thanks Kit, that does the trick!

... though having to add ,xlab(,format(%td)) kills the simplicity of the lfitci approach, so I think I'll stay with my second attempt, where I can also control the significance level.


The data has high kurtosis so I'm thinking of using something other than the t-distribution for the confidence intervals, does that make sense? I'm thinking of replacing invttail with some other inverse tail, does it make any sense?


Patrick.



my original question below:

Dear statalist,

A very simple question: I want to plot confidence intervals onto a time series variable.

The problem I have is that in some cases the dates display as dates while in other cases they display as numbers. I would like the dates to display as dates.

I used rates.dta as an example, from the web data repository, even though it doesn't produce interesting confidence intervals. Feel free to use another dataset for illustration.

The following is my test code. Attempt 1 is what I understand to be the natural approach: the graph has problems with the dates in this case. Attempt 2 is my attempt to do things "by hand": the dates display as expected in this case.

Follow-up question 1: is there a better way to plot confidence intervals with a flexible value of the significance level, than my second attempt below?

Follow-up question 2: can I plot the confidence intervals for the logarithm of the variable "on the fly" or do I need to define a new variable equal to the log of the level variable?

Follow-up question 3: does stata provide a confidence intervals command under non-normal assumptions (e.g. Pareto)? or do I need to do it by hand along the lines of my second attempt below?

I suspect my questions are FAQ so please do feel free to simply direct me to some relevant url.

many thanks.


 clear
 cap log close
 set more off
 webuse rates.dta

 global x1 dow
 format date %td
 tsset date
 global t date

 /* Construct a 95 percent confidence interval, attempt 1 */
 graph twoway (lfitci $x1 $t) ///
              (scatter $x1 $t) ///
			  , title("95% Confidence Interval (dates wrong!)")
 graph export ci95_x1.ps,replace

 /* Construct a 95 percent confidence interval, attempt 2 */
 reg $x1 $t, robust
 predict yhat, xb
 predict shat, stdp
 scalar define alpha = 0.9 /*significance level*/
 gen crit = invttail(e(df_r),alpha/2)
 gen upper = yhat + crit*shat
 gen lower = yhat - crit*shat
 graph twoway (scatter $x1 $t) ///
              (line yhat $t) ///
			  (line lower $t) ///
			  (line upper date) ///
              , title("Confidence Interval, significance level = `=alpha'")
 graph export ci_x1.ps,replace
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index