14
Common Empirical Methods and Stata Jared DeLisle

Common Empirical Methods and Stata Jared DeLisle

Embed Size (px)

DESCRIPTION

Common Empirical Methods and Stata Jared DeLisle. Topics We Will Cover. Regression (OLS), adjustment of standard errors, and output Sorting firms by characteristic(s) Portfolio returns based on a strategy Matching firms by characteristic(s) Calendar-time portfolios - PowerPoint PPT Presentation

Citation preview

Page 1: Common Empirical Methods and  Stata Jared DeLisle

Common Empirical Methods and Stata

Jared DeLisle

Page 2: Common Empirical Methods and  Stata Jared DeLisle

Topics We Will Cover

Regression (OLS), adjustment of standard errors, and output

Sorting firms by characteristic(s)

Portfolio returns based on a strategy

Matching firms by characteristic(s)

Calendar-time portfolios

Fama-MacBeth (1973) regressions

Page 3: Common Empirical Methods and  Stata Jared DeLisle

Resources

-Stata linkshttp://www.personal.psu.edu/fpv/sourcecode.htmhttp://www.eszter.com/stata.htmlhttp://www.ats.ucla.edu/stat/stata/http://personal.anderson.ucla.edu/judson.caskey/data.htmlhttp://dss.princeton.edu/online_help/stats_packages/stata/http://people.su.se/~mkuda/stata.htmlhttp://ideas.repec.org/s/boc/bocode.html

http://www.kellogg.northwestern.edu/faculty/petersen/htm/papers/se/se_programming.htm

-SAS Linkshttp://www.ats.ucla.edu/stat/sas/http://dss.princeton.edu/online_help/stats_packages/sas/Boehmer, Broussard, & Kallunki, Using SAS in Financial Research

-Remember, Google is your friend!

Page 4: Common Empirical Methods and  Stata Jared DeLisle

Stata 11 (current version)

Stata is a statistical package which runs on Windows, Macintosh and Unix platforms.- Just like SAS, it is powerful, however, IMHO, much easier to use than SAS.

- For a comparison of programs, visit: http://www.ats.ucla.edu/stat/technicalreports/number1_editedFeb_2_2007/ucla_ATSstat_tr1_1.1_0207.pdf

Page 5: Common Empirical Methods and  Stata Jared DeLisle

Stata 11

Review Window- All entered commands are listed for easy recall

Variables Window- Lists variables (and labels) contained in dataset

Command Window- Commands are entered here one line at a time

Output Window- Results, and errors, show up here

Page 6: Common Empirical Methods and  Stata Jared DeLisle

Stata 11

Page 7: Common Empirical Methods and  Stata Jared DeLisle

Stata 11

Data can be accessed and analyzed using the command line prompt- Review pane makes debugging and recall easy

Commands can also be input via a “do-file”- “Do-files” are text files with the .do file extension- Allows Stata to run through a series of commands easily and maintains reproducibility of analysis

Page 8: Common Empirical Methods and  Stata Jared DeLisle

Regression

OLS- regress depvar indepvars

OLS with White (or robust) standard errors- regress depvar indepvars, r- regress depvar indepvars, cluster(var)

OLS with fixed effects- areg depvar indepvars, absorb(var) (use r or cluster() for robust se’s)

Probit (Logit, Tobit) regression- probit (logit) depvar indepvars (use ,r or cluster() for robust se’s)

2-D Clustered standard errors (Petersen [2009])- cluster2 depvar indepvars, tcluster(time) fcluster(firm)- probit2, logit2, tobit2

Page 9: Common Empirical Methods and  Stata Jared DeLisle

Sorting

“xtile” (or, alternatively, “xtileJ” [J. Caskey]) sorts sample into the number of groups you specify by the variable you specify- xtile newvar = var, nquantiles(#)- xtileJ newvar = var, nquantiles(#) by(byvar)

Page 10: Common Empirical Methods and  Stata Jared DeLisle

Portfolio Returns

Given a strategy, a researcher wishes to learn if the strategy produces abnormal returns.

In order to do this, the researcher can only use information that an investor would have at the time of portfolio formation, and then examine the portfolio returns in the next period.

Typically involves a “long-short” or “zero-cost” portfolio

Let’s do an example with momentum returns…

Page 11: Common Empirical Methods and  Stata Jared DeLisle

Calendar Time Portfolios

Typically for long-run studies, when the researcher wishes to form a portfolio of firms where an event triggers the firm’s entrance into the portfolio and the firm stays in portfolio for a certain amount of time (12 months, 36 months, etc.). Fama (1998) recommends this method over buy-and-hold abnormal returns.

Let’s look at how we might create a dataset with calendar time portfolio returns and analyze such a dataset…

Page 12: Common Empirical Methods and  Stata Jared DeLisle

Exercise – Matching firms

There are various methods (ranging from simple to sophisticated to ridiculous) of matching a firm in a sample to a firm out-of-sample.

Let’s think about how we could set up a simple match between our in and out-of-sample firms based on Fama and French (1997) 48-industries, size, and 12-month momentum.

Page 13: Common Empirical Methods and  Stata Jared DeLisle

Fama-MacBeth (1973) Regressions

Very common in asset pricing

First, time series regressions on each group’s returns to estimate factor betas

Followed by cross-sectional regressions each time period to estimate risk premiums on the factor betas

The estimated risk premiums are averaged over all time periods, and se’s are calculated

xtfmb, xtfmbJ (J. Caskey), fmtest (R. Tharyan)

Page 14: Common Empirical Methods and  Stata Jared DeLisle

Some things we didn’t cover

Event studies (for those without access to Eventus, see http://dss.princeton.edu/online_help/stats_packages/stata/eventstudy.html)

GMM, Heckman, Simultaneous equations, IV regression Time-Series analysis

- VAR, VECM, ARIMA, GARCH, Unit-root and cointegration tests, etc.

Survey analysis Hazard analysis Maximum Likelihood Estimation (Non-linear, FIML, LIML, etc.) Other stuff we can’t possibly cover in 1 hour Good news! Stata (& SAS) can do most of these analyses!

- Again, Google is your friend!... And so am I!

[email protected]