How to loop variables to run regression with one variable

[Tan Nguyen]

Hi all, I have sample datasets as follow:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(month BLF BMC BMP mkt mkt_lag mkt_lead)
600          .          .           .          .          . -.08011035
600 -.07186875 -.13838592 -.073767476 -.08011035          . -.10704234
600 -.14840126 -.14784347  -.10364176 -.10704234 -.08011035 -.11275777
600 -.00837866 -.14142856  -.14624734 -.11275777 -.10704234 -.13514103
600 -.07071709    -.11154   -.1559959 -.13514103 -.11275777          .
600          .          .           .          . -.13514103 -.14756103
600 -.17334187  -.1534142   -.1457479 -.14756103          . -.09404086
600 -.09202663 -.10659858   -.1224689 -.09404086 -.14756103 -.10542256
600 -.12763597    -.11271  -.06582951 -.10542256 -.09404086 -.12629919
600  -.1667886  -.1528849     -.11272 -.12629919 -.10542256 -.14956874
600 -.16150147 -.16314194   -.1516216 -.14956874 -.12629919  -.0945335
600 -.05489313 -.10300398   -.0803421  -.0945335 -.14956874          .
600          .          .           .          .  -.0945335 -.13635546
600  -.1302676 -.14886476   -.0977102 -.13635546          .          .
600          .          .           .          . -.13635546 -.10551248
600  -.1040715 -.09425955     -.11244 -.10551248          .          .
600          .          .           .          . -.10551248 -.13636737
600  -.1535125 -.07911568   -.1522368 -.13636737          . -.12262583
600 -.13812657 -.16431424  -.14554626 -.12262583 -.13636737 -.11261494
600 -.12154094    -.11303     -.11303 -.11261494 -.12262583 -.10477968
601   -.123664    -.11508  -.06667894 -.10477968 -.11261494  -.1130095
601 -.12367833    -.11502  -.08401432  -.1130095 -.10477968 -.09727497
601 -.09538767 -.09524008  -.09747676 -.09727497  -.1130095 -.09626808
601 -.10446905 -.10439398  -.10548863 -.09626808 -.09727497  -.1345686
601 -.12143094 -.14770897   -.1432283  -.1345686 -.09626808 -.11931893
601 -.13282233    -.11558   -.1468279 -.11931893  -.1345686          .
601          .          .           .          . -.11931893 -.09770807
601 -.09616494 -.09615672  -.07464554 -.09770807          .          .
601          .          .           .          . -.09770807          .
601          .          .           .          .          . -.10804106
601  -.1056115 -.10532606     -.11398 -.10804106          .  -.1412695
601 -.14788257  -.1580387  -.15229504  -.1412695 -.10804106  -.1201113
601   -.108096 -.10770897     -.11668  -.1201113  -.1412695          .
601          .          .           .          .  -.1201113 -.11151864
601 -.08119761 -.10657697   -.1311398 -.11151864          . -.10317887
602 -.07422082 -.09799007   -.1233489 -.10317887 -.11151864 -.11929395
602 -.12338744 -.12394395  -.12332292 -.11929395 -.10317887 -.10276413
602 -.09917016     -.1153  -.10726708 -.10276413 -.11929395          .
602          .          .           .          . -.10276413          .
602          .          .           .          .          . -.09910545
602  -.0754342  -.0887317   -.0828021 -.09910545          .  -.1028476
602 -.09210663    -.11458    -.099311  -.1028476 -.09910545  -.1184791
602    -.11544 -.12381052    -.130709  -.1184791  -.1028476 -.12176133
602 -.14987817     -.1198  -.12751527 -.12176133  -.1184791  -.1045903
602 -.12526308     -.1176  -.07214075  -.1045903 -.12176133 -.11538172
602 -.10837692 -.07488298     -.11604 -.11538172  -.1045903 -.14801851
602  -.1647848  -.1674981  -.16335924 -.14801851 -.11538172  -.1280758
602 -.14218338    -.11789  -.13351673  -.1280758 -.14801851 -.10276656
602 -.08773822 -.10318842  -.10437327 -.10276656  -.1280758 -.12722206
602 -.12796839  -.1453183  -.13562673 -.12722206 -.10276656 -.12638341
602  -.1503241     -.1178  -.14170773 -.12638341 -.12722206 -.12992042
602 -.11176926       -.12  -.12809797 -.12992042 -.12638341 -.10705458
602 -.09373662 -.12661602  -.10993203 -.10705458 -.12992042 -.13630375
602 -.15163413 -.13649993  -.15190294 -.13630375 -.10705458  -.1149129
602 -.11046926     -.1187      -.1187  -.1149129 -.13630375          .
602          .          .           .          .  -.1149129 -.12884723
602 -.11049905  -.1445438  -.13581052 -.12884723          . -.12180012
602 -.12735094 -.07664081     -.11884 -.12180012 -.12884723 -.10072123
603 -.09315144 -.09397837  -.10166947 -.10072123 -.12180012 -.11403426
603 -.13535762 -.10254543  -.12691091 -.11403426 -.10072123  -.1090324
603  -.0841027 -.13344458  -.13439108  -.1090324 -.11403426 -.11777158
603 -.14371818 -.14331163  -.10186892 -.11777158  -.1090324  -.1178696
603 -.12725912 -.12711486  -.12725963  -.1178696 -.11777158 -.11860677
603 -.09371343    -.11882  -.11038037 -.11860677  -.1178696          .
603          .          .           .          . -.11860677 -.11403073
       .
end
format %tm month

I want to run regression of each firm(e.g BLF,....) on excess market return (mkt mkt_lag mkt_lead) for one-month window, but i realise that I toke me a lot of time when i do as follow:

bys month: asreg BLF mkt mkt_lag mkt_lead
drop _Nobs- _adjR2 _b_cons
gen beta_BLF = _b_mkt + _b_mkt_lag + _b_mkt_lead
drop _b_mkt - _b_mkt_lead

bys month: asreg BMC mkt mkt_lag mkt_lead
drop _Nobs- _adjR2 _b_cons
gen beta_BMC = _b_mkt + _b_mkt_lag + _b_mkt_lead
drop _b_mkt - _b_mkt_lead

bys month: asreg BMP mkt mkt_lag mkt_lead
drop _Nobs- _adjR2 _b_cons
gen beta_BMP = _b_mkt + _b_mkt_lag + _b_mkt_lead
drop _b_mkt - _b_mkt_lead

I would appreciate any suggestions how can I approach this problem because i have more than 600 firms and do such things spend a lot of my time. Thank you very much.

[Clyde Schechter]

The difficulty arises because you have the data in wide layout. Once you reshape it to long, it's just a few lines of code:

gen long obs_no = _n

ds obs_no month mkt*, not
local firms `r(varlist)'

rename (`firms') x=

reshape long x, i(obs_no) j(firm) string
rangestat (reg) x mkt mkt_lag mkt_lead, by(firm) interval(month . .)
gen beta = b_mkt + b_mkt_lag + b_mkt_lead
drop reg_nobs reg_adj_r2 b_mkt* se_*

Note: -rangestat- is by Robert Picard, Nick Cox, and Roberto Ferrer. It is available from SSC. I do not use -asreg-, so I cannot be sure, but I suspect you can use -asreg- instead of -rangestsat- here as well in some way if you are more comfortable with that. That would also require adjusting the subsequent two lines to reflect the variable naming conventions in -asreg-.

Note 2: I dropped the standard errors. If you want them, you can modify the -drop- command.

Note 3: The -reshape long- may take a while. You need to be patient with a large data set. If it really becomes unreasonable, consider getting Rafal Raciborski's -tolong- from SSC, or Mauricio Caceres Bravo's -gtools- package, which includes a -greshape- command. Both of them run considerably faster than Stata's native -reshape- in large data sets.

Note 4: While you probably should keep the data in long layout, as that will make almost anything you want to do after this easier than with wide data, if you must return to wide layout for some reason, you can do that as follows:

rename (reg_r2 b_cons beta) =_
reshape wide x reg_r2_ b_cons_ beta_, i(obs_no) j(firm) string