Final assignment (group)
Do this project in groups of 2 people (or if necessary I'll allow one
group of three people). There are two deliverable: a short paper and
a class presentation. The content is flexible, I'll suggest some
ideas at the end--but they are only suggestions. You can pick your
own.
-
Class presentation (last two days of class): You and your
partner should give a 10-15
minute in-class presentation of your results. It isn't
necessary for each of you to talk, but feel free to do it that
way if you like.
- You MUST have a one page handout. It is always
important to send
your audience home with something that has your name on it (and
in the real world you should also provide contact
information. Otherwise, how can they hire
you to consult?)
- You may use overheads, chalk, power point or even
JMP. But since we don't want to see how you
run your numbers, JMP would only be useful for
"brushing" or other interactive statistics.
- Your primary audience is your fellow students, not me.
Putting them to sleep isn't nice!
- Write-up: The write-up is due before the end of classes. It might
be useful to do the write-up first and give me a copy. (I'm
willing to even look over rough drafts or outlines.) If you do
that, I will email you your comments (if you provide me with an
email address of one of your group members) so you will have a
chance to fix things before the presentation. Assume it will
take me 48 hours to read and comment.
- Think of this as the usual discussion from previous
homeworks.
- Keep it short! Aim for 2 pages of text and a maximum of 5
pages of appendix. The idea is that the text should carry
the primary economic story. It should have graphs that are important to
the primary message. For example, you wouldn't put
a histogram of the residuals in the main text--instead put
it in the appendix. The
appendix can include other
graphs that
support your story. You don't have to have an appendix if
don't need one.
- Put captions on your figures. Make your figures simple
and clear.
- If you turn in early, I'll allow a revision as I have on
earlier homeworks. If you
turn in on the last day of classes, your document is
final. I won't force any resubmits. So turning early
will not be held against you.
Project ideas
The following is a list of project ideas. It is more of a starting point
for discussion with your partner rather than a complete list of
possibilities. The basic idea is that you should find some data that
is similar to something discussed in class and present it. By
similar, I mean statisticially similar rather than substantially
similar.
The focus should be on what we learn from the data (namely what
interesting economic fact did we learn from the data) and not on the
statistical methodology.
- Learning curve example: Find data on computer chip manufacturing. You
need both the price and the total amount produced. You should
take a quick look at Ray Kurzweil's new book. You will need
to learn enough about chip manufacturing to understand the
process and describe what might have been learned.
- Production function: Find data on the number of lines of
computer code written by a group of programmers. As you add
programmers, their efficiency goes down--so you should find a
classic diminishing marginal return to programmer. If you can
also find out information on number of managers, you can have
two terms in your production function.
- Ponzironi data: Look at some successful scheme--be it a mutual
fund, a hedge fund, a trading method based on the past 200 day
rolling average, or a particular trader (say
turtletrader.com). Come to a decision as to whether true
excess returns are generated or not. (You will need to read a
prospectus or other information to determine what a disaster
would be. You will need to think hard about how much search
you (and others) did to find this scheme.)
- Ponzironi simulation: Look at real instruments (puts, calls,
small high volatility stocks, etc) available to say a mutual fund
or to a hedge fund. Find a way of converting these instruments
into a bet in which you mostly win a little bit, and sometimes
lose a lot. If you can find historical data on your
instrument, run a "look back" analysis of how well your fund
would have performed over the past years. (Say 10/20 years for
a mutual fund and 1-5 years for a hedge fund.)
- Time series (evaluations): Find some historical predictions
(say from the Economist which often puts forecasts on its last
page) of a macro-economic variable (say GNP, interest rates,
unemployment, etc). Collect 10-20
years of data. Now you have both a forecast and the actual
result and so can compute a RMSE. Try a VERY simple model and
compute its RMSE. How do the experts do compared to this
simple model?
- Time series (forecasting): Decide on some macro-economic
variable to predict (say interest rates, unemployment, GNP,
etc). Think of at least 2 other variables that might be
significant drivers of your variable of interest. Build a
model on data from 1980-1989 using your variables. DO NOT look
at the data from 1990-2000. After you have a model that you
are happy with generate predictions for 1990-2000. Compute
your RMSE on these 10 years of data. Did you over fit?
- Marketing: Find some data relating advertising (or other
promotion) to sales. For example, Wharton has scanner data
from grocery stores available from WRDS.