The time period of the data is from Jan 2, 2004 to the present.
"sparta" represents the daily close prices of the Spartan International Index Fund.
"fworld" represents the daily close prices of the Fidelity Worldwide Fund. This is a similar mutual fund and is included to control for external market shocks that effected international mutual funds in general.
"Prod" takes the value of 1 for the dates that fall within the month that 300 began production.
"Comic" that takes the value of 1 for the dates that fall within the month that ComicCon occurred. There was a large-scale advertisement for 300 at ComicCon.
"myspace" takes the value of 1 for the dates that fall within the month that the 300 Myspace was opened.
"release" takes the value of 1 for the dates that fall within the month or following month of the movie release.
"postrelease" takes the value of 1 for the third month following the release.
"dvd" takes a value of 1 for the dates that fall within the month of the DVD release.
Both regressions are heteroskedastic, so I ran them using robust standard errors.
The results from the first regression suggest that these promotions of 300 correlate with price increases of the Spartan International Index Fund. Call me crazy, but this might indicate the people got excited about 300 and were influenced to invest more money in the Spartan International Index Fund, causing the price to increase.
The results of the second regression suggest that all promotions of 300 correlate with price decreases of the Fidelity Worldwide Fund, which is a close substitute of the Spartan International Index Fund. This may be the effect of investors allocating more funds toward the Spartan and, therefore, less to substitutes.
This plot represents the Spartan fund's prices over the time of the sample. The first line marks ComicCon, the second Myspace, and the third the release.
So, correlation doesn't prove causation, but what about six? Does anyone else interpret this as evidence that investment choices can be influenced by pop-culture?
I'm not sure what the connection is exactly, but I would be naive to think there is not a connection given this data.
5 comments:
This is a flawed regression model.
Your r-squared values are around .98 for both of your ill conceived models. That means that you've successfully accounted for 98% of the variance in the prices for both of these funds with your variables?!!?
Assuming your model is correct, which it so obviously is not, means that your crazy ass variables are what truly determine stock price, and the other 2%...you know things like rates of return, tax rates, etc, are hardly influential.
The main problem with your model, aside from the fact that you built a model around 7 dummy variables, is that your variables are multi co-linear up the wazoo. Every single variable you have chosen is a measurement of the passing of time, directly or indirectly.
Now, without even running a regression, I'm sure we can witness from your data a positive correlation between the price of the Spartan fund and time. Yet, all the variables you've chosen are just round about ways of examining this trend.
Your model effectively shows that an investment fund, whose price is increasing over time, has a positive correlation with the passing of time.
You've completely failed to explain why the price is rising. The only reason why your correlation isn't a perfect 1, I suspect, is because some of your dummy variable time intervals are overlapping.
Your second regression model is flawed in the exact same ways.
I'll give you the benefit of the doubt, and take this tongue and cheek regression analysis as the joke you've intended it as. However, building a model around multi-colinearity is hardly humorous. This joke could've been made just as well without improperly using a regression analysis.
I dearly hope that you were aware of the flaws in this model before you published this, however.
First, thanks for your comment. I don't get many and I really appreciate them.
Second, I get the feeling I'm talking to someone a lot smarter than me, but allow me a brief chance to defend my joke.
The r-squared values are high because fworld is included in the sparta model and visa versa. So, it's not that my crazy dummy variables explain more than rate of return, tax rates, etc. Rather, most of the shocks that affect one international mutual fund also affect other international mutual funds. All prices correlate.
Also, there are only six dummy variables. The first variable is fworld, which I include to control for shocks common to other international mutual funds over time. So, even though the other six are just pieces of time, they measure how one fund price changed during each time period controlling for the change in the other fund's price.
If each fund were not included in the other's model, the early (late) dummy variables would be negative (positive) because, like you said, the fund's price increased over time.
That aside, the model is flawed in a number of other ways.
First, I control for the price of the Fidelity Worldwide Index, but this choice of fund is arbitrary. Picking another fund could change the results. Second, the dates of all the dummy variables are arbitrary. Changing the dates would likely change the results. The list of flaws is long...
The joke that I had in mind was just the idea that mutual fund prices might be affected by the titles and promotions of blockbuster films. I mean, the image of a bunch of office workers buying mutual funds and screaming, "This is Sparta!" ...I thought it was funny.
But you're right. There's nothing funny about multicollinearity and I apologize.
| sparta fworld prod comic myspace release postre~e dvd
-------------+------------------------------------------------------------------------
sparta | 1.0000
fworld | 0.9921 1.0000
prod | -0.0733 -0.0824 1.0000
comic | 0.0228 0.0029 -0.0212 1.0000
myspace | 0.1351 0.1144 -0.0212 -0.0207 1.0000
release | 0.2439 0.2093 -0.0303 -0.0296 -0.0296 1.0000
postrelease | 0.2145 0.1997 -0.0217 -0.0212 -0.0212 -0.0303 1.0000
dvd | 0.1795 0.1842 -0.0217 -0.0212 -0.0212 -0.0303 -0.0217 1.0000
You're right. There is a multicollinearity problem. Give me time. I will improve the joke and post again.
Post a Comment