Error: unknown function ()

alessio lombini

Join Date: Dec 2020

Posts: 98
#1

Error: unknown function ()

21 Jul 2021, 03:32

Hello everyone,

I am trying to compute tariff equivalent effects following the approach suggested in An Advanced Guide to Trade Policy Analysis. page 30 of the book. I attach below the code proposed in the book and the error I get when I try the second line of the code:

Code:

*** Model for GVC participation asdoc ppmlhdfe gvc_total regulatory_distance contig applied_tariff ln_dist comlang_off comlang_ethno comcol comrelig col45 comleg_pretrans sibling rta if exp!=imp, a(exp_time imp_time) vce(cluster pair_id) replace * Tariff equivalent effects scalar TariffEquivalentRD_1 = (exp(_b[regulatory_distance]/(-_b[applied_tariff])) – 1) * 100 unknown function () r(133);

Does someone know how to fix this issue?

Thanks in advance
Tags: None
Carlo Lazzaro

Join Date: Apr 2014

Posts: 17352
#2

21 Jul 2021, 03:47

Alessio:
have you already double-checked number and order of brackets?

Kind regards,
Carlo
(StataNow 18.5)
Comment
alessio lombini

Join Date: Dec 2020

Posts: 98
#3

21 Jul 2021, 06:30

Dear Carlo,
thank you very much for the prompt reply. Yes, I have checked the number of brackets and it is not wrong. Also, I tried to delete the first pair of parenthesis and I got the following issue:

Code:

. scalar TariffEquivalentRD_1 = exp(_b[regulatory_distance]/(-_b[applied_tariff] > )) – 1 invalid '–'
Comment
Andrew Musau

Join Date: Oct 2014

Posts: 9638
#4

21 Jul 2021, 06:48

I guess that you can write the expression as follows:

Code:

scalar TariffEquivalentRD_1 = (`=exp(_b[regulatory_distance]/(-_b[applied_tariff])) -1')*100

It appears that scalar is unable to directly evaluate the expression as written.

Last edited by Andrew Musau; 21 Jul 2021, 06:57.
Comment
alessio lombini

Join Date: Dec 2020

Posts: 98
#5

21 Jul 2021, 07:26

Dear Andrew,

thank you very much for your help. I followed your suggestion and the code as you proposed seems to work. However, I am not sure about what I get. Normally, it should yield a positive or negative percentage. I tried it with different models and I always get this result. Is it possible that there is an error?

Code:

. di TariffEquivalentRD_1 -100

Thank again for your support
Comment

Andrew Musau

Join Date: Oct 2014
Posts: 9638

21 Jul 2021, 09:24

In performing arithmetic operations, order matters, hence the acronyms PEMDAS (Parenthesis Exponents Multiplication Division Addition Subtraction), BODMAS (Brackets Orders Division Multiplication Addition Subtraction), BIDMAS (Brackets Indices Division Multiplication Addition Subtraction), etc. All these imply the same order of operations. So you can verify whether you are getting a sensible result based on how you expect the expression to be evaluated. Therefore, if you want to substract before multiplying, then parenthesize the elements to be subtracted. Substituting regulatory_distance with "weight" and applied_tariff with "trunk", I get

Code:

sysuse auto, clear
regress mpg weight trunk
scalar test = (`=exp(_b[weight]/(-_b[trunk])) -1')*100
di test

Res.:

Code:

. regress mpg weight trunk

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(2, 71)        =     67.19
       Model |  1598.76953         2  799.384767   Prob > F        =    0.0000
    Residual |  844.689926        71  11.8970412   R-squared       =    0.6543
-------------+----------------------------------   Adj R-squared   =    0.6446
       Total |  2443.45946        73  33.4720474   Root MSE        =    3.4492

------------------------------------------------------------------------------
         mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      weight |  -.0056527   .0007016    -8.06   0.000    -.0070516   -.0042537
       trunk |   -.096229   .1274771    -0.75   0.453    -.3504112    .1579531
       _cons |   39.68913    1.65207    24.02   0.000       36.395    42.98327
------------------------------------------------------------------------------



. scalar test = (`=exp(_b[weight]/(-_b[trunk])) -1')*100

.
. di test
-5.7049911

which seems to be right as

[exp(-.0056527/0.096229)]- 1 = -.05705014 and (-.05705014*100)\(\approx\) -5.705.

Last edited by Andrew Musau; 21 Jul 2021, 09:33.

Comment

alessio lombini

Join Date: Dec 2020
Posts: 98

21 Jul 2021, 09:55

Dear Andrew,
thank you very much for your detailed and comprehensive explanation. See the code attached below:

Code:

 regress gvc_total regulatory_distance applied_tariff

      Source |       SS           df       MS      Number of obs   =   582,015
-------------+----------------------------------   F(2, 582012)    =    167.21
       Model |  70713332.2         2  35356666.1   Prob > F        =    0.0000
    Residual |  1.2307e+11   582,012  211454.026   R-squared       =    0.0006
-------------+----------------------------------   Adj R-squared   =    0.0006
       Total |  1.2314e+11   582,014  211574.797   Root MSE        =    459.84

-------------------------------------------------------------------------------
    gvc_total |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
regulatory_~e |   39.87936   7.115612     5.60   0.000     25.93299    53.82573
applied_tar~f |  -1.098885   .0603646   -18.20   0.000    -1.217198   -.9805724
        _cons |    56.6861   1.004809    56.41   0.000      54.7167    58.65549
-------------------------------------------------------------------------------

. 
end of do-file

. do "C:\Users\Utente\AppData\Local\Temp\STD4478_000000.tmp"

. scalar TariffEquivalentRD = (`=exp(_b[regulatory_distance]/(-_b[applied_tariff
> ])) -1') * 100

. 
end of do-file

. do "C:\Users\Utente\AppData\Local\Temp\STD4478_000000.tmp"

. di TariffEquivalentRD
5.766e+17

. 
end of do-file

. do "C:\Users\Utente\AppData\Local\Temp\STD4478_000000.tmp"

. regress gvc_total regulatory_distance applied_tariff

      Source |       SS           df       MS      Number of obs   =   582,015
-------------+----------------------------------   F(2, 582012)    =    167.21
       Model |  70713332.2         2  35356666.1   Prob > F        =    0.0000
    Residual |  1.2307e+11   582,012  211454.026   R-squared       =    0.0006
-------------+----------------------------------   Adj R-squared   =    0.0006
       Total |  1.2314e+11   582,014  211574.797   Root MSE        =    459.84

-------------------------------------------------------------------------------
    gvc_total |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
regulatory_~e |   39.87936   7.115612     5.60   0.000     25.93299    53.82573
applied_tar~f |  -1.098885   .0603646   -18.20   0.000    -1.217198   -.9805724
        _cons |    56.6861   1.004809    56.41   0.000      54.7167    58.65549
-------------------------------------------------------------------------------

. 
end of do-file

. do "C:\Users\Utente\AppData\Local\Temp\STD4478_000000.tmp"

. scalar TariffEquivalentRD = (`=exp(_b[regulatory_distance]/(-_b[applied_tariff
> ])) -1') * 100

. di TariffEquivalentRD
5.766e+17

. 
end of do-file

. do "C:\Users\Utente\AppData\Local\Temp\STD4478_000000.tmp"

. qui asdoc ppmlhdfe gvc_total regulatory_distance contig applied_tariff ln_dist
>  comlang_off comlang_ethno comcol comrelig col45 comleg_pretrans sibling rta i
> f exp!=imp, a(exp_time imp_time industry) vce(cluster pair_id) replace

. scalar TariffEquivalentRD = (`=exp(_b[regulatory_distance]/(-_b[applied_tariff
> ])) -1') * 100


. di TariffEquivalentRD
-100

Might it be possible that the issue is related to the model ppmlhdfe ?
Thanks again for your time!

Comment

Andrew Musau

Join Date: Oct 2014

Posts: 9638
#8

21 Jul 2021, 10:28

Look at the coefficients

Code:

di _b[regulatory_distance] di _b[applied_tariff]

If the numerator is large relative to the denominator, taking the exponential of the ratio will result in a value close to or equal to 0. Then you will observe your result, e.g.,

Code:

di exp(-1000/ 1.2)

Res.:

Code:

. di exp(-1000/ 1.2) 0

Last edited by Andrew Musau; 21 Jul 2021, 10:54.
Comment
alessio lombini

Join Date: Dec 2020

Posts: 98
#9

23 Jul 2021, 09:56

Dear Andrew,
I am afraid you are right, the numerator is indeed large compared to the numerator. I guess I cannot really use the estimates I get and I cannot overcome this issue.
Thank you very much for your help
Comment

Andrew Musau

Join Date: Oct 2014
Posts: 9638

#10

23 Jul 2021, 11:18

You can change the magnitude of a coefficient by rescaling the variable. This changes nothing except the units of the variable.

Code:

sysuse auto, clear
regress mpg trunk weight
replace weight= weight/1000
regress mpg trunk weight

Res.:

Code:

. regress mpg trunk weight

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(2, 71)        =     67.19
       Model |  1598.76953         2  799.384767   Prob > F        =    0.0000
    Residual |  844.689926        71  11.8970412   R-squared       =    0.6543
-------------+----------------------------------   Adj R-squared   =    0.6446
       Total |  2443.45946        73  33.4720474   Root MSE        =    3.4492

------------------------------------------------------------------------------
         mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       trunk |   -.096229   .1274771    -0.75   0.453    -.3504112    .1579531
      weight |  -.0056527   .0007016    -8.06   0.000    -.0070516   -.0042537
       _cons |   39.68913    1.65207    24.02   0.000       36.395    42.98327
------------------------------------------------------------------------------

.
. replace weight= weight/1000
variable weight was int now float
(74 real changes made)

.
. regress mpg trunk weight

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(2, 71)        =     67.19
       Model |  1598.76956         2  799.384782   Prob > F        =    0.0000
    Residual |  844.689895        71  11.8970408   R-squared       =    0.6543
-------------+----------------------------------   Adj R-squared   =    0.6446
       Total |  2443.45946        73  33.4720474   Root MSE        =    3.4492

------------------------------------------------------------------------------
         mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       trunk |   -.096229   .1274771    -0.75   0.453    -.3504112    .1579531
      weight |  -5.652679   .7015899    -8.06   0.000     -7.05161   -4.253749
       _cons |   39.68913    1.65207    24.02   0.000       36.395    42.98327
------------------------------------------------------------------------------

.

Announcement