Re: st: Implementing a Momentum Investment Strategy with Stata

[Robson Glasscock <glasscockrc@vcu.edu>]

Daniel,
I agree with Richard. His advice is much better than mine. I was
suggesting a manual approach instead of using -egen, xtile-. The
identifier variable would mark the top 10% of performers in each
window. If you know each stock in each respective window, then you can
write a statement that is conditional on each window and would mark
the subsequent returns for these stocks during the following 6 months.
These are the returns you would then use to calculate the return on
the momentum portfolios.

All apologies for the rabbit trail,
Robson Glasscock


On Tue, Mar 6, 2012 at 2:10 PM, Richard Herron
<richard.c.herron@gmail.com> wrote:
> You're right it is -egenmore- from SSC.
>
> The -egen, xtile- command should generate a variable -portfolio- with
> integers 1 through 10. When there is no return for that month -egen,
> xtile- won't be able to assign a portfolio.
>
> The -f6.-, -l6.-, and -s6.- are time series operators that shift
> forward, lag, or seasonally difference by 6 observations. Before you
> can use them you will need to tell Stata your panel identifiers with
> -tsset- or -xtset-.
>
> I don't understand your return notation. You should just have one
> variable that contains the n month returns for all stocks. After you
> -tsset- your data (something like -tsset permno date_ym-) then the
> code from my first reply should work.
>
> On Tue, Mar 6, 2012 at 10:49,  <schmani@gmx.de> wrote:
>> Dear Robson, dear Richard,
>>
>> thank you very much for your input!
>> I haven't yet decided whose approach I am gonna follow, hence I have some more issues for clarification:
>>
>> Robson:
>> By introducing an "identifier" variable do you mean that I need to have one variable for each time-window and hence e.g. tag the top 10 performers with a "1"? How can I achieve this when I only have 385 observations, yet the before-mentioned 1400 stocks? I'm not sure I got the intuition behind your idea. Could you kindly clarify this for me?
>>
>> Richard:
>> I assume you meant the egenmore package from ssc, since I could not find egemore? You are right that Jegadeesh and Titman used equally weighted portfolios, this will also be the approach I follow for my study. I tried using your egen command for one sample-portfolio. Is it correct that the generated variable "portfolio" now has only 1s in it for the time period where the stock has returns?
>> I probably need to fully understand what you were trying to achieve with your usage of f6., l6. and f6. Could you kindly elaborate on that? I should have mentioned that I have log-returns (named: return_v#-return_v####) whose observations consist of the cumulated previous 6m returns where available.
>> It seems to me that the command generates a portfolio based on only one asset (ret6), not on a portfolio of several assets.
>>
>> (my command looked like this: egen portfolio = xtile( return_v1377), nquantiles(10) by (date)  ) Do I have to use the whole set of data (i.e. return_v#-return_v####) instead of just one return set?
>>
>> Your help is greatly appreciated, as I am quite a novice in Stata.
>>
>> Best regards,
>>
>> Daniel
>>
>>
>> -------- Original-Nachricht --------
>>> Datum: Sun, 4 Mar 2012 17:40:42 -0500
>>> Von: Richard Herron <richard.c.herron@gmail.com>
>>> An: statalist@hsphsun2.harvard.edu
>>> Betreff: Re: st: Implementing a Momentum Investment Strategy with Stata
>>
>>> I don't think there's a need to use -rolling-. I think there are two
>>> main steps here. Generating the portfolios, then generating the time
>>> series of returns for eash of these portfolios.
>>>
>>> I would generate the portfolios using -xtile- from the -egemore- package
>>> (SSC).
>>>
>>> generate ret6 = s6.price / l6.price
>>> egen portfolio = xtile(ret6), nquantiles(10) by(date_ym)
>>>
>>> where -ret6- is your holding period return and -date_ym- is the
>>> year-month date (say 2001m5). Now generate the portfolio holding
>>> period returns using collapse and time series operators.
>>>
>>> collapse (mean) f6.ret6, by(date_ym portfolio)
>>>
>>> You may also want to use weightings to determine portfolio weights
>>> (although I think Jegadeesh and Titman is all equally weighted). Now
>>> you can use these portfolios to generate returns to your strategy.
>>> HTH.
>>>
>>> On Sun, Mar 4, 2012 at 09:20, Robson Glasscock <glasscockrc@vcu.edu>
>>> wrote:
>>> > Christopher Baum's "An Introduction to Stata Programming" discusses
>>> > moving-windows in Chapter 8. The -rolling- command is mentioned as
>>> > well as Cox and Baum's -mvsumm- and -mvcorr- programs (available from
>>> > SSC). Looking into these may help you think about a modified approach
>>> > to your problem.
>>> >
>>> > Instead of using matrices, I wonder if it would be better to generate
>>> > two new variables that would help you identify the top 10% of firms in
>>> > each window. First, generate a "window" variable that would take on a
>>> > value of 1 for months 1-6, 2 for months 2-7, 3 for months 3-8, etc.
>>> > Second, generate an "identifier" variable that tags the top 10% of
>>> > firms in each window. You would know exactly which firms were included
>>> > in each window and then it would be possible for you to calculate the
>>> > returns for those firms over the next six months. Or maybe clever use
>>> > of -rolling- is the better approach here. I'm not sure, but I hope
>>> > this is enough to get you started.
>>> >
>>> > best,
>>> > Robson Glasscock
>>> >
>>> >
>>> > On Sun, Mar 4, 2012 at 7:43 AM,  <schmani@gmx.de> wrote:
>>> >> Dear all,
>>> >>
>>> >> I am trying to implement a Momentum Strategy following
>>> Jegadeesh/Titman(1993): Returns to Buying Winners and Selling Losers: Implications for
>>> Stock Market Efficiency, JoF 48(1), pp. 65-91.
>>> >>
>>> >> I have monthly returns for ~1400 stocks for a total of 385 time periods
>>> and now have to proceed in the following way:
>>> >>
>>> >> 1) build cumulated returns over a formation period of 6months (I did
>>> that by generating new variables)
>>> >>
>>> >> 2) select the top 10% performing stocks in order to invest. This is
>>> done every month in order to generate so-called "overlapping" portfolios.
>>> >>
>>> >> 3) invest in those stocks for a holding period of 6 months
>>> >>
>>> >> 4) determine the average of the returns and adjust for risk
>>> >>
>>> >>
>>> >> For now, my approach was to generate matrices for every month t that
>>> consist of the cumulated returns over the previous 6 months.
>>> >> However I can not find an efficient way to sort these matrices while
>>> being able to identify which assets to invest in (this is crucial as I need
>>> to tell Stata somehow which assets to invest in).
>>> >>
>>> >> I am sure there is a better way to sort the cumulated returns of the
>>> previous 6 months in every month t while still being able to somehow "track"
>>> the variable name in order to invest in the corresponding stock.
>>> >> I hope I made myself clear, any help would be greatly appreciated!
>>> >>
>>> >> Best,
>>> >> Daniel
>>> >> --
>>> >> NEU: FreePhone 3-fach-Flat mit kostenlosem Smartphone!
>>> >> Jetzt informieren: http://mobile.1und1.de/?ac=OM.PW.PW003K20328T7073a
>>> >> *
>>> >> *   For searches and help try:
>>> >> *   http://www.stata.com/help.cgi?search
>>> >> *   http://www.stata.com/support/statalist/faq
>>> >> *   http://www.ats.ucla.edu/stat/stata/
>>> >
>>> > *
>>> > *   For searches and help try:
>>> > *   http://www.stata.com/help.cgi?search
>>> > *   http://www.stata.com/support/statalist/faq
>>> > *   http://www.ats.ucla.edu/stat/stata/
>>>
>>> *
>>> *   For searches and help try:
>>> *   http://www.stata.com/help.cgi?search
>>> *   http://www.stata.com/support/statalist/faq
>>> *   http://www.ats.ucla.edu/stat/stata/
>>
>> --
>> NEU: FreePhone 3-fach-Flat mit kostenlosem Smartphone!
>> Jetzt informieren: http://mobile.1und1.de/?ac=OM.PW.PW003K20328T7073a
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/statalist/faq
>> *   http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/