Ivo Welch and Matthew Rothman, Econ 1759, Fall 2008

Economics 1759, Fall 2008, is an experimental course that will teach modern research techniques in financial economics. The skills are useful both in graduate economics courses and in quantitative Wall Street research departments, and specifically those that work on quantitative equities investing. (Matthew Rothman is the global head of quantitative equity strategy research at Lehman Bros, and holds a PhD from the University of Chicago.)

Homeworks

1. Done. (Matching GVKEYs and PERMNOs.)
2. Done. (Creating Market-Model Regression Estimates (betas, alphas, sigmas). Reading single compustat items.)
3. Producing a complete data set for analysis with betas and specific Compustat fields.
4. Running a specific Fama-Macbeth study. Running a specific Fama-French sort study.
5. Running an open-ended study.

Chronological News and/or Announcements

2008/12/05

The final project, which is what your class grade is based on, consists of three parts, using the data set that you have put together. The main specification was:

Returns <- Lagged Firm Size, Lagged Book/Market, Lagged Accruals, 1-mo Momentum, 2-7mo Momentum, Market-Beta

1. Run a Fama-Macbeth regression predicting returns with these 6 variables. (Note: you do not have to do the multistep Fama-Macbeth beta pre-sorting and portfolio formation procedures. You can use the historical betas and individual firms.) Consider how your inference differs in Januaries vs. non-Januaries, and for small firms vs. large firms (based on ex-ante firm size).
2. Use other variables from your data set to see if you can beat this model.
3. Run a Fama-French time-series regression. You should form portfolios using dependent double-sort techniques. That is, for each of these variables, first sort by firm-size, then spread (within groups of 4) by each variable. Then compute the rates of return for this portfolio in time-series. Finally, use the FF factors to determine the alpha.

When you hand in a written report that details this work (say, 5 pages of writeup and 5 pages of tables), you receive a grade. In your report, talk about both statistical and economic significance (think "how many basis points" meaning). The project is due on the last day of the reading period. If you cannot finish the project in time, and you want to receive a grade in this class, you must talk to us before the deadline.

2008/12/05

Class Plan: I plan to talk about the research project that I have been working on for the last 2 years.

2008/11/14

• HW 3a: create the data set and submit Dec 2004 by Tuesday eve. HW 3b: run a Fama-Macbeth regression, as specified, by Friday class.
• Clarification: When you compute the rate of return for a month, you must start compounding with the first rate of return in this month (even though [or, better, because] it is partly based on the price at the end of the previous month. Also, a stock that delists in the middle of the month must have a monthly rate of return in this month, even though it is based on only some of the returns in this month. (This is to avoid throwing out firms that may go bankrupt and delist mid-month.)
• For reference checking purposes, Alex has prepared one line for a version of homework 3 that you can use for checking purposes. pdf, and xlsx.
• Please be prepared to talk about Matt's research on Friday; please be prepared to show off your program that executes homework 3. This should be completed by today.
• It looks like I will get permission to count a good report with a good homework 4 (an intelligent, self-designed trading strategy implementation with a written report) as the equivalent of a senior thesis that obtains an honors annotation on the transcript. The deadline for delivery of a complete draft of such a homework 4 will be January 31, 2009. (If I approve it, we should be able to count it as an honor's project. Do not start work on a strategy before you talk to me; you might be disappointed if you show me only a completed version that I then would not approve.)

2008/11/10

Please be prepared to talk about Matt's research on Friday; please be prepared to show off your program that executes homework 3. This should absolutely be completed by this Friday.

2008/11/03

From Matthew: wsresearch.

2008/10/30

Data Set To Create:

• Plan for 2008/10/31

2008/10/28

Today's Clarifications:

• Compustat does not provide a date when the company released the annual statement, or a date when it entered the data into its data base. (Yikes! How stupid.)
• We will only use annual data. We will assume that the entry date of annual statement information into Compustat is 5 months after the fiscal year has ended, so we can trade on it 6 months after the fact. (This is conservative. 4 months would often do, too.
• Here is the mapping file from crsp to compustat and vice-versa that you should use for homework 3 and beyond: crsp_compustat_mapping.csv. Discard anything else that you may have used earlier.
• You must be running your full alpha/beta/stderrs/stddev market-model regression program now, so that you have all this information at week's end in one big file.

2008/10/24

Today's Clarifications:

• Store the beta information for all firm-months in a separate file. Suggestion: use the format:

    permno mdate alpha beta sealpha sebeta sigmasq
  

  where sigmasq is the variance of the residuals.
• Ignore all records that do not have one unique GVKEY to PERMNO match. A reference file for your convenience is going to be posted here.
• Ignore all firms that do not have 5 years of continuous data for computing beta. (Save yourself the time.)
• Write out "NaN" for missing data. All smart regression packages know that this indicates a missing value.
• Make sure your data pull from Compustat does not use restarted numbers, but original numbers; and store also the date on which the information was released (or entered the data base).

(I plan to give a few lines of correct examples for homework 3.)

2008/10/24

Matt's presentation: Anomalies.

2008/10/23

Please make sure to bring your notebook computer to class tomorrow, and be prepared to explain where you are with respects to the assignments, what works, and what does not work. Alex and I are planning to do short individual consulting sessions.

2008/10/10

Due to equity market turbulence, Matt cannot come today, and I am in TX. Thus, today's class is cancelled. Don't slack off: First, if you have not handed in homework 2, please send an email to Alex asap and tell him where you are stuck, and how he/we can help you to become unstuck. Also, please start on homework 3. We want to be close to done with it by the end of next week.

2008/10/08

Matt would like you to run a cross-sectional regression of stock return on the estimated betas. I guess the easiest way to do this for returns from Jan/1/2005 to Feb/1/2005 since you have already calculated betas applicable for this date. Try to do this by Friday if you can.

Ex-Post

Next year, have hw 2a, which computes beta reg info for all firm months.

2008/10/05

Please turn in homework 2. Put the following files into the homework directory on the data server:

• An R or Matlab or like program, called hw2.R (or equivalent).
• A perl, ruby, python or like program, called hw2.pl (or equivalent).
• Provide us the output (hw2.csv): the set of market-betas for all firms as of 12/31/2004. That is, your very last rate of return that may be included in the beta calculation is that on 12/31/2004, because we would want to start trading on it on the morning of 1/1/2005. Your returns should be based on 252*5 trading days. Make sure you have the dating right.

  More details: For the market and risk-free rates of return, let's switch to Ken French's data in F-F_Research_Data_Factors_daily.zip, That is, please do not use the S&P500, but Ken's "Mkt-RF" column; and as your own stock's rate of return in the market model, please use it net of the "RF" rate. (Obviously, this may require small change in your program if you used the S&P500 earlier.) Your output format should be a csv file with two columns, PERMNO, BETA. We want to post the results by Friday, so please post your results by Tuesday Noon. (Honor Code Note: You are not allowed to look at the programs and results of your colleagues. Yes, there is a Unix log of file accesses.)

2008/10/05

You should start on homework 3 now. Homework 3 requires you to write a program that that can take a set of Compustat variables and create a data set that looks as follows:

permno gvkey yyyymm this-months-stock-return-from-CRSP ...a set-of-lagged-and-or-computed-variables-from-Compustat and your betas...

Your program must have a reasonably easy way to allow you to change the lagging parameter and the variables that you want in your data set. The output should be gzipped. For example, I may tell you that I want you to use 3-month lagged variables, (except the own rate of return, of course), and that I want [a] market-betas; [b] Net-Income/Sales (i.e., net-income and sales may come from a financial statement that is between 3 to 3+11 months old); and [c] Sales. [Sales may be called Revenues.] A typical line may therefore look as follows:

10045,231465,200511,0.123,1.241,0.054,921341.3
10045,231465,200512,-0.123,1.229,0.054,921341.3

and the last two items may be 6 months old. My guess is that for 40 years of data, should have about 40*12*5,000∼2-3 million lines of data. Once the data is gzipped, it should be a few megabytes. (Extra-Credit: Think about whether it would be difficult to do YTD versions of your compustat variables.)

2008/10/03

homework 1 results

2008/09/29

The corrected BKM Table 13.8.

2008/9/12

We will have a guest speaker November 9.

Older Notes

See oldernotes.html.

Summaries of Old Classes

2008/10/03

Plan

2008/09/26

Plan

2008/09/19

Notes. Plan.

2008/09/12

Plan.

2008/09/05

Notes

Data

CRSP

CRSP's programming interface is essentially unusable. Therefore, we have prepared simple data sets for you. On the data server, please connect to /data/crsp2008/ and copy the files in this directory to your own notebook. The '.txt.gz' files can be uncompressed with 'zcat *.txt.gz'. Their formats are very simple: PERMNO (the CRSP identifier), tab, the data piece, tab, and the day. The contents of the 'crsp_names.txt' files are described in the header. Please read the CRSP documentation at FILLIN.

Compustat

Please refer to Alex's website.

Yahoo

Some sample program for translating yahoo quote.csv finance data into yhoo2csv.pl.

Preliminary Syllabus

The draft and abbreviated version of the syllabus is in at data-course2.html and possibly data-course2.pdf.