9:40 p.m.
On Fri, 26 Dec 2008, John Paravantis wrote:
How about adding the option to draw asymptotic confidence
intervals when a nonlinear model is NOT autoregressive?
OK, added in CVS and Windows snapshot.
Allin.
9:16 a.m.
New subject: cannot forecast with Nonlinear Least Squares
Thank you very much Allin!
Apropos, a very happy new year to you and all gretl users
John Paravantis
University of Piraeus
GREECE
Allin Cottrell wrote:
On Fri, 26 Dec 2008, John Paravantis wrote:
> How about adding the option to draw asymptotic confidence
> intervals when a nonlinear model is NOT autoregressive?
>
OK, added in CVS and Windows snapshot.
Allin.
9:34 a.m.
New subject: fitted actual values over time problem in CVS
Allin,
There may exist a problem with the current CVS version that includes the
confidence intervals.
With the following data set (attached for you to receive in your
personal email)
obs INDEX CARS100 CARS100_132 CARS100_083
1970 1 2.58038212 4.14090472 3.79424228
1971 2 3.00538458 4.44290805 4.10112574
1972 3 3.44037669 4.76611495 4.43143460
1973 4 3.84414443 5.11189066 4.78672465
1974 5 4.35824937 5.48166985 5.16861656
1975 6 4.85366685 5.87695734 5.57879199
1976 7 5.56997899 6.29932832 6.01898816
1977 8 6.68061911 6.75042793 6.49099094
1978 9 7.76015994 7.23197026 6.99662599
1979 10 8.64942636 7.74573642 7.53774806
1980 11 8.92886330 8.29357183 8.11622797
1981 12 9.56137905 8.87738238 8.73393733
1982 13 10.20066268 9.49912947 9.39273078
1983 14 10.88537085 10.16082373 10.09442560
1984 15 11.59668762 10.86451737 10.84077867
1985 16 12.69542033 11.61229486 11.63346086
1986 17 13.59994140 12.40626198 12.47402880
1987 18 14.27350165 13.24853305 13.36389430
1988 19 14.96199763 14.14121615 14.30429175
1989 20 15.90075064 15.08639646 15.29624375
1990 21 17.11635187 16.08611733 16.34052572
1991 22 17.44837205 17.14235940 17.43762996
1992 23 17.86531335 18.25701752 18.58773016
1993 24 19.02780635 19.43187567 19.79064719
1994 25 20.03533056 20.66857995 21.04581723
1995 26 21.16687196 21.96860986 22.35226342
1996 27 22.29307030 23.33324805 23.70857224
1997 28 23.61610553 24.76354911 25.11287575
1998 29 25.05426070 26.26030746 26.56284087
1999 30 27.19582718 27.82402528 28.05566675
2000 31 29.43956697 29.45488073 29.58809106
2001 32 31.35982081 31.15269740 31.15640588
2002 33 33.28769299 32.91691552 32.75648341
2003 34 34.97170795 34.74656600 34.38381158
2004 35 37.04166841 36.64024791 36.03353921
2005 36 39.08502957 38.59611033 37.70052974
2006 37 NA 40.61183951 39.37942272
2007 38 NA 42.68465197 41.06470122
2008 39 NA 44.81129435 42.75076371
2009 40 NA 46.98805043 44.43199823
2010 41 NA 49.21075601 46.10285688
2011 42 NA 51.47482150 47.75792854
2012 43 NA 53.77526256 49.39200772
2013 44 NA 56.10673834 51.00015785
2014 45 NA 58.46359691 52.57776732
2015 46 NA 60.83992724 54.12059711
2016 47 NA 63.22961660 55.62481913
2017 48 NA 65.62641249 57.08704470
2018 49 NA 68.02398755 58.50434320
2019 50 NA 70.41600620 59.87425092
2020 51 NA 72.79619144 61.19477088
2021 52 NA 75.15839033 62.46436428
2022 53 NA 77.49663676 63.68193457
2023 54 NA 79.80521012 64.84680537
2024 55 NA 82.07868865 65.95869328
2025 56 NA 84.31199658 67.01767691
2026 57 NA 86.50044423 68.02416320
2027 58 NA 88.63976044 68.97885218
2028 59 NA 90.72611731 69.88270109
2029 60 NA 92.75614678 70.73688873
2030 61 NA 94.72694969 71.54278078
2031 62 NA 96.63609726 72.30189663
2032 63 NA 98.48162587 73.01587821
2033 64 NA 100.26202550 73.68646111
2034 65 NA 101.97622290 74.31544818
2035 66 NA 103.62356003 74.90468578
2036 67 NA 105.20376888 75.45604271
2037 68 NA 106.71694327 75.97139164
2038 69 NA 108.16350866 76.45259321
2039 70 NA 109.54419069 76.90148236
2040 71 NA 110.85998297 77.31985704
2041 72 NA 112.11211507 77.70946881
2042 73 NA 113.30202099 78.07201544
2043 74 NA 114.43130861 78.40913508
2044 75 NA 115.50173060 78.72240194
2045 76 NA 116.51515699 79.01332327
2046 77 NA 117.47354955 79.28333747
2047 78 NA 118.37893825 79.53381314
2048 79 NA 119.23339981 79.76604893
2049 80 NA 120.03903832 79.98127413
2050 81 NA 120.79796801 80.18064975
(only variable CAR100 matters) try to run nonlinear least squares with
the following script
genr N0=5
genr K=130
genr r0=0.1
CARS100=N0*K/(N0+(K-N0)*exp(-r0*INDEX))
params N0 K r0
When the NLS model window opens, select the menu Graphs>Fitter, actual
plot>Against time. A graph window opens momentarity (unusually large,
compared to customary gretl plots) and then you get the following dialog box
Failed to open file 'C:\Documents and Settings\John
Paravantis\Application Data\gretl\gretltmp.png': No such file or
directory
after which you get another similar dialog box. When you click OK in the
second dialog box the graph closes!
Incidentally, all other graphs plot OK.
John
John Paravantis wrote:
Thank you very much Allin!
Apropos, a very happy new year to you and all gretl users
John Paravantis
University of Piraeus
GREECE
Allin Cottrell wrote:
> On Fri, 26 Dec 2008, John Paravantis wrote:
>> How about adding the option to draw asymptotic confidence
>> intervals when a nonlinear model is NOT autoregressive?
> OK, added in CVS and Windows snapshot.
>
> Allin.
3:56 p.m.
New subject: fitted actual values over time problem in CVS
On Tue, 30 Dec 2008, John Paravantis wrote:
There may exist a problem with the current CVS version that
includes the confidence intervals...
When the NLS model window opens, select the menu Graphs>Fitter,
actual plot>Against time...
Urgh, yes, and get a big mess! This is now fixed in CVS and the
Windows snapshot. (This was a newly introduced bug, and nothing
specifically to do with nonlinear least squares -- it would have
messed up any fitted vs. actual plot -- so thanks for spotting
it!)
A little more on what I've done with NLS and forecasting. We now
check whether or not an NLS equation is dynamic (i.e. contains one
or more lags of the dependent variable on the right-hand side).
If so, we offer a dynamic forecast out of sample, but we do not
produce any forecast standard errors / confidence intervals, since
I don't know how we could produce defensible values without
bootstrapping (which would require a fair amount of work).
If the model is _not_ dynamic (as in your cars example), we
produce forecast standard errors as simply the standard error of
the NLS regression (abstracting from parameter uncertainty), and
confidence intervals that are point forecast plus/minus t(df,
0.025) * SE.
Allin.
8:08 p.m.
New subject: fitted actual values over time problem in CVS
Allin, thanks for the info.
As far as I am concerned, I would very much like it if you added
bootstrapping capabilities to gretl so that asymptotic confidence
intervals could be plotted.
I clearly understand that this involves quite a lot of programming, so
feel free to prioritize it as you see appropriate.
Best,
John
Allin Cottrell wrote:
On Tue, 30 Dec 2008, John Paravantis wrote:
> There may exist a problem with the current CVS version that
> includes the confidence intervals...
>
> When the NLS model window opens, select the menu Graphs>Fitter,
> actual plot>Against time...
>
Urgh, yes, and get a big mess! This is now fixed in CVS and the
Windows snapshot. (This was a newly introduced bug, and nothing
specifically to do with nonlinear least squares -- it would have
messed up any fitted vs. actual plot -- so thanks for spotting
it!)
A little more on what I've done with NLS and forecasting. We now
check whether or not an NLS equation is dynamic (i.e. contains one
or more lags of the dependent variable on the right-hand side).
If so, we offer a dynamic forecast out of sample, but we do not
produce any forecast standard errors / confidence intervals, since
I don't know how we could produce defensible values without
bootstrapping (which would require a fair amount of work).
If the model is _not_ dynamic (as in your cars example), we
produce forecast standard errors as simply the standard error of
the NLS regression (abstracting from parameter uncertainty), and
confidence intervals that are point forecast plus/minus t(df,
0.025) * SE.
Allin.
12:18 a.m.
New subject: fitted actual values over time problem in CVS
On Tue, 30 Dec 2008, John Paravantis wrote:
Allin, thanks for the info.
As far as I am concerned, I would very much like it if you added
bootstrapping capabilities to gretl so that asymptotic confidence
intervals could be plotted.
John, what exactly do you mean by "asymptotic confidence
intervals" in the context of NLS? Can you give a reference?
Bootstrapping would presumably not produce such intervals; they
would be empirical rather than based on asymptotic theory.
Allin.
12:44 a.m.
New subject: fitted actual values over time problem in CVS
On Tue, 30 Dec 2008, Allin Cottrell wrote:
On Tue, 30 Dec 2008, John Paravantis wrote:
> As far as I am concerned, I would very much like it if you added
> bootstrapping capabilities to gretl so that asymptotic confidence
> intervals could be plotted.
John, what exactly do you mean by "asymptotic confidence
intervals" in the context of NLS? Can you give a reference?
Bootstrapping would presumably not produce such intervals; they
would be empirical rather than based on asymptotic theory.
I should perhaps add: I was talking about the possibility of
bootstrapping specifically in the context of _dynamic_ NLS models,
where the standard error of the regression is clearly an
underestimate of the uncertainty of a dynamic forecast, even in
the limit where (let's suppose) parameter uncertainty goes to nil.
Are you talking about dynamic or static models?
Allin.
8:54 a.m.
New subject: confidence intervals in NLS
Allin,
Please see PRISM 4 manual that may be downloaded for free from
http://www.graphpad.com/manuals/prism4/regressionbook.pdf
In it (pp. 97 to 103), you will find references to asymptotic standard
errors and confidence intervals in the case of STATIC models estimated
via NLS.
I have also attached the NCSS curve fitting manual to this message (to
be received in your private email). In it you will find references to
bootstrapping for confidence intervals.
To me, it seems that both manuals refer to the SAME thing.
I was referring to the possibility of adding similar functionality to
gretl, but please keep in mind that I am a USER of econometrics - I
certainly do not have a PhD on the subject.
John Paravantis
Allin Cottrell wrote:
On Tue, 30 Dec 2008, Allin Cottrell wrote:
> On Tue, 30 Dec 2008, John Paravantis wrote:
>
>
>> As far as I am concerned, I would very much like it if you added
>> bootstrapping capabilities to gretl so that asymptotic confidence
>> intervals could be plotted.
>>
> John, what exactly do you mean by "asymptotic confidence
> intervals" in the context of NLS? Can you give a reference?
> Bootstrapping would presumably not produce such intervals; they
> would be empirical rather than based on asymptotic theory.
>
I should perhaps add: I was talking about the possibility of
bootstrapping specifically in the context of _dynamic_ NLS models,
where the standard error of the regression is clearly an
underestimate of the uncertainty of a dynamic forecast, even in
the limit where (let's suppose) parameter uncertainty goes to nil.
Are you talking about dynamic or static models?
Allin.
3:44 p.m.
New subject: confidence intervals in NLS
On Wed, 31 Dec 2008, John Paravantis wrote:
Please see PRISM 4 manual that may be downloaded for free from
http://www.graphpad.com/manuals/prism4/regressionbook.pdf
In it (pp. 97 to 103), you will find references to asymptotic standard
errors and confidence intervals in the case of STATIC models estimated
via NLS.
Thanks, but that discussion pertains to standard errors and
confidence intervals for the parameter estimates -- something
gretl already does fine -- not for forecasts.
I have also attached the NCSS curve fitting manual to this
message (to be received in your private email). In it you will
find references to bootstrapping for confidence intervals.
Yes, the section on "Bootstrap prediction intervals" in that text
is relevant. This is something we can probably implement at some
point.
Allin.
5866
days inactive
5868
days old
8 comments
2 participants
participants (2)
-
Allin Cottrell -
John Paravantis