[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Data Interpolation

From   "Bromiley, Philip" <[email protected]>
To   <[email protected]>
Subject   st: Data Interpolation
Date   Wed, 23 Jan 2008 12:39:53 -0800

How bad interpolation is may depend on what one intends to do with the
variables.  For example, if you have an event that occurs in a couple of
months, then you really need monthly data on the dependent variable and
the explanatory variables whose coefficients you want to interpret.  If
I'm looking at say monthly exchange rates, particularly in a period when
you have abnormal changes, I can't depend on interpolation for the
factors that spark such changes.  It simply cannot have the relevant
information (e.g., what happened in June that caused a change in July).
Interpolating may also introduce spurious correlations.

However, you might have some control variables which change little
within a year where interpolation would be less dangerous.  Population
or US consumer prices might be useful controls in a model with many
years of monthly data.


Philip Bromiley
University of California, Irvine
Irvine, CA 92697-3125
(949) 824-6657
(949) 725-2898 (fax)
Date: Tue, 22 Jan 2008 11:16:19 +0000 (GMT)
From: Maarten buis <[email protected]>
Subject: st: Re: Data Interpolation

- --- Ismail Ait Saadi <[email protected]> wrote:
> Thank you again for your reply. Well as I told you the analysis has
> no value using ANNUAL DATA, I am not the first one to do so. My area
> is currency crises and top economists in the field use monthly data,
> because it capture the sudden nature of currency crises. 

How can you hope to capture something if you haven't seen it? The fact
that smart people have done it is no argument, even smart people can do
stupid things. At the very least this confirms to common conception
that economist don't do research: all they do is imagine what the world
looks like. 

All this sounds harsh (and part of it is tongue in cheek). All I want
to do is challange you to consider where your conclussions come from:
If you are interested is some sudden change, and your data is too
coarse to see it, than any conclussion can not come from your
observations. If it does not come from your observations than it comes
from your assumptions (which untill now I have called your

> Yes I understand your worries but what to do? I have about 5
> indicators that should be interpolated. I tried Kit suggestion but it
> didn't work, maybe because he didn't give me a good example to
> follow. If you can clarify it please do if you have time.

reg gdp monthlyvars time
predict gdp_monthly

Here the monthlyvars are the variables you do observe on a monthly
basis, time is time but could also be a nonlinear function of time like
time and time squared or a spline (see: -help mkspline-). Your hope
would be that the crisis is picked up in the monthlyvars, and thus
transmitted to your new gdp_monthly. Appart from finding the right
model for time, another thing to think about is whether some of these
montlyvars need to be lagged and if so by how much...

I also repeat that if you continue along this road you should take a
good long look at multiple imputation.

- -- Maarten

- -----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index