NOTES ON PROJECTS

The project this semester involves estimating a multiple
regression model, where the CANSIM number for the dependent
variable will be assigned to you in tutorial.  The period
for which the data are claimed to be available is noted.
You must determine which independent variables (generally,
at least three variables are needed) to use to "explain" the
dependent variable, find the data on these independent
variables using CANSIM, download all the data, estimate the
resultant equation, do some diagnostic testing, and write up
the results.  Although some of the topics assigned can be
viewed as esoteric, similar problems are often given to
persons working in the business or economic consulting
fields.

A link to the CANSIM web page is available on my web page,
and after accessing that you should click on "search
CANSIM".  For the dependent variable (where you know the
CANSIM number), you can then click on "retrieve a single
cansim series by label", and once retrieved (use spreadsheet
format), save it to your diskette using the "save" command
in the "file" menu.  To find series for which you do not
have the CANSIM number, click on  "search cansim catalog
files".  If the matrix search command does not yield what
you want try the "main index" or "full text index" options.
You may also want to try the "access the alphabetical list
of matrices" button.  Don't give up easily - there are a lot
of series in CANSIM, but finding what you want is NOT
(usually) easy.  You should try using CANSIM early in the
semester - don't wait until week 11 to find out how it
works.

Note that before entering data into EXCEL, all series must
have the same start and end date.  This can be achieved in
two different ways:  (1) when getting the series from
CANSIM, ensure that all variables you will want are
available for a common period, and use this period to set
beginning and ending dates on the retrieve command, or (2)
edit the data after it has been retrieved, e.g. in WORD.

The summary report (5 pages maximum) should note (i) a brief
theoretical or ad hoc rationale for why the independent
variables selected should influence the dependent variable,
including expected signs, (ii) the exact specification of
the independent variables, including any adjustments made to
the data, problems of missing observations or
transformations made, (iii) any changes to the sample period
or periodicity of the data (e.g. monthly to quarterly) due
to problems of obtaining desired data, (iv) assessment of
regression diagnostics, including looking at the residuals
to determine outliers and testing for autocorrelation
(section 14.8), and (v) an interpretation of the results.
Computer output, with the relevant results highlighted, must
be turned in with the report for grading.

I strongly suggest people check with their TA once they have
determined which independent variables they think they
should try to find, to make sure they have not omitted
something important.  I suggest you use the maximum period
for which data are available for all variables.  You will
often find that the independent variables you have chosen
will not be available for the full period for which the
dependent variable is available.  It is permissible to
shorten the observation period as long as at least 20
observations are used for annual data, 30 observations for
quarterly data, or 50 observations for monthly data.  Do not
shorten the data period more than is necessary, as a loss of
power (in the statistical sense) will result, and the amount
of work you have to do is not reduced by shortening the
period in any event.  If it appears this sample size "rule"
must be violated, please check with your TA before
proceeding.  I suggest using the data with the periodicity
given for the dependent variable, if necessary changing the
periodicity of the independent variables to match (e.g. you
can "make" monthly data out of quarterly data by repeating
each observation three times).  On the other hand, if you
have a lot of observations of monthly data on your dependent
variable, and some independent variable is only available in
quarterly form, it may be desirable to change all variables
to a quarterly basis.  This can be done on the retrieve
command in CANSIM.  If in doubt, check with your TA.

Finally, it should be noted that I did not actually download
most of the data or check that there was as much data
available as the header indicated.  You may find that the
data contain less observations than indicated, or that more
recent observations have become available since I looked
(CANSIM is updated weekly).  Further, I did note in passing
that some series contain missing data points.  I suggest you
fill these in by interpolation, noting on your writeup that
you did so.