Originally posted by the blog Calculated Risk and then duplicated on The Big Picture, the figure below shows employment numbers for various recessions in U.S. history. Obviously this data is relevant to many people at the moment. Anybody in my PSTAT 262 care to take a stab at this data as part of their project?

I am curious why the data seem so jumpy in earlier years relative to the smooth curve for the current recession. Bad data gathering techniques? Untrustworthy news sources led to businesses not reacting at the same time?

The plot below from the National Data Buoy Center is from two different stations (no idea where they are located) measuring air pressure. Although the data shown in this plot isn’t current, data of this sort would be appropriate for my PSTAT 262 project. Apparently the upshot of this particular figure was to decide that the buoy at 42043 has a bias in its reporting of air pressure since it is consistently above the air pressure at station 42035. Any reason to think the air pressure might just be consistently higher there? I have no idea, so I guess I will take the analyst word for it.

Here are a number of datasets that could possibly be used for a project in PSTAT 262. All the time series indicate, in some fashion, our current economic condition.

Previously I have mentioned a plan to have students analyze a data set.  Here is an interesting dataset concerning housing starts. From wikipedia, housing starts “refers to the number of privately owned new homes (technically housing units) on which construction has been started in a given period.” This index gives an indication of how the housing market is doing.

Ever since watching An inconvenient truth, I have wondered about this figure that shows an apparent relationship between carbon dioxide and global temperature.  I couldn’t find a better version of this picture.

Previously I announced a plan to have students in my PSTAT 262 class model and forecast actual time series. Here is another example of a possible dataset for modeling which concerns US Debt from (almost) all sources as a percentage of GDP. As mentioned here, this debt does not include a few trillion in “off balance sheet” financing, contingent unfunded pension plans for corporate and state and local governments, or unfunded liabilities of the U.S. government for such items as Medicare, Social Security and other programs.

As mentioned in a previous post, I will be asking students in my PSTAT 262 class to choose and analyze a data set. I thought posting some suggestions on this blog would be a good way to introduce some possibilities.

Shown below is a time series found at Barry Ritholtz’s blog which tracks the cost of moving goods via the sea.

As part my of PSTAT 262 (Applied Bayesian Time Series) next semester, I plan on having students pick out a data set to make both non-model and model based forecasts for the series. My only requirement for the data are that the correct forecasts cannot already exist.  An additional suggestion is that the forecasts come true soon so that forecasts can be compared with realizations.

Examples of unacceptable datasets are

• Dow Jones Index from 1990-2000
• Dollars spent on US health care until 2005
• Abundance of passenger pigeons

Examples of acceptable datasets are

• Yearly average S&P 500 Index from 1949 to present
• Mean global temperature in the past millennium
• Median U.S. House price for the past 40 years

Examples of ideal datasets are

• Daily price of Coca-Cola stock since 2000
• Number of daily airline passengers for the past 5 years
• Weekly sunspot activity since 1977

Students will be asked to forecast their time series at two different time points: a relatively short forecast and one relatively long.

Here is a post by a student in EEP100 at Berkeley on the aguanomics blog. This is the first of a series of posts by students.

I’m considering implementing this for my PSTAT 262MC class on Applied Bayesian Time Series. My thought is to require students to post twice during the quarter. The first post will try to forecast a time series using just intuition, i.e. no math/stats. The second post will try to forecast a time series (possibly the same one) using a statistical model based approach.

Any thoughts?