8
$\begingroup$

I have a dataset with 20 products and whether someone bought them, their price and other attributes.

I am trying to find the own and cross-price elasticities of these 20 goods. I have used a conditional logistic model in Stata and now need to find the elasticities preferably in a matrix. I am thinking of using margins or mfx. Using mfx would I need to use the below? Or is there a better approach to finding the price elasticity matrix in Stata?

mfx predict(bought = 1) dyex(ln(product = '1'))
mfx predict(bought = 2) dyex(ln(product = '2'))
etc.

Regression results were generated as per:

 clogit bought productX attrY for X = 1..20 and Y = 1..15.

Could I use mfx eyex as to get the own and cross-price elasticities?

$\endgroup$
  • 2
    $\begingroup$ Can you provide the code how you fit the conditional logit? $\endgroup$ – Andy May 24 '14 at 12:46
  • 2
    $\begingroup$ mfx was superseded by margins, so you should probably be using the latter. You can benefit from: Stata tip 88 by Christopher Baum, Stata Journal (2010). $\endgroup$ – Roberto Ferrer May 24 '14 at 12:54
  • $\begingroup$ It seems that mentioned results related to marginal effects and not elasticity. Elasticity and marginal effects seems to be different. One tries to get changes over ONE unit and the other ONE Percent. $\endgroup$ – user46777 Jun 5 '14 at 11:52
7
$\begingroup$

To estimate the matrix of own- and cross-price elasticities it is more convenient to use the asclogit command in Stata. Let's use the example in Cameron and Trivedi (2009) to show you the whole process to get this matrix. The data is on individual choices of whether to fish from the beach, the pier, a private boat or a chartered boat. They have three variables:

  • income is each individual's income
  • p is the price of each fishing alternative
  • q is the quantity of fish that an individual can obtain from each fishing alternative

Note that income is individual specific (a consumer characteristic) and p and q are choice specific (the product characteristics). Now let's first estimate the choice model and then obtain the matrix of own- and cross-price elasticities.

Obtain the data set and prepare it for the regression:

net from http://www.stata-press.com/data/musr
net install musr
net get musr
use mus15data.dta
reshape long d p q, i(id) j(fishmode beach pier private charter) string

Run the alternative specific clogit (asclogit) model:

asclogit d p q, case(id) alternatives(fishmode) casevar(income) basealternative(beach) nolog

Alternative-specific conditional logit         Number of obs      =       4728
Case variable: id                              Number of cases    =       1182

Alternative variable: fishmode                 Alts per case: min =          4
                                                              avg =        4.0
                                                              max =          4

                                                  Wald chi2(5)    =     252.98
Log likelihood = -1215.1376                       Prob > chi2     =     0.0000

------------------------------------------------------------------------------
           d |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
fishmode     |
           p |  -.0251166   .0017317   -14.50   0.000    -.0285106   -.0217225
           q |    .357782   .1097733     3.26   0.001     .1426302    .5729337
-------------+----------------------------------------------------------------
beach        |  (base alternative)
-------------+----------------------------------------------------------------
charter      |
      income |  -.0332917   .0503409    -0.66   0.508     -.131958    .0653745
       _cons |   1.694366   .2240506     7.56   0.000     1.255235    2.133497
-------------+----------------------------------------------------------------
pier         |
      income |  -.1275771   .0506395    -2.52   0.012    -.2268288   -.0283255
       _cons |   .7779593   .2204939     3.53   0.000     .3457992    1.210119
-------------+----------------------------------------------------------------
private      |
      income |   .0894398   .0500671     1.79   0.074    -.0086898    .1875694
       _cons |   .5272788   .2227927     2.37   0.018     .0906132    .9639444
------------------------------------------------------------------------------

A negative coefficient for the price means that if the price of one fishing choice increases, then the demand for this choice decreases and the demand for the other choices increases. Conversely, if the quantity of fish caught increases in one choice, the demand for this choice increases and decreases for the other choices.

Now the own- and cross price elasticities are $$ \begin{align} \frac{\partial p_{ij}}{\partial x_{rik}} &= p_{ij}(1-p_{ij})\beta_r, \qquad j=k \quad\text{(own-price elasticity)} \newline \frac{\partial p_{ij}}{\partial x_{rik}} &= -p_{ij}p_{ik}\beta_r, \quad \quad \quad j \neq k \quad \text{(cross-price elasticity)} \end{align} $$ for the alternative choice $r$ (relative to the baseline choice; here the baseline is "beach").

The own- and cross-price elasticities are easily estimate as:

estat mfx, varlist(p)

Pr(choice = beach|1 selected) = .05248806
-------------------------------------------------------------------------------
variable     |   dp/dx   Std. Err.    z     P>|z|  [    95% C.I.    ]       X
-------------+-----------------------------------------------------------------
p            |                                                                 
       beach | -.001249   .000121  -10.29   0.000  -.001487  -.001011    103.42
     charter |  .000609   .000061    9.97   0.000   .000489   .000729    84.379
        pier |  .000087   .000016    5.42   0.000   .000055   .000118    103.42
     private |  .000553   .000056    9.88   0.000   .000443   .000663    55.257
-------------------------------------------------------------------------------

Pr(choice = charter|1 selected) = .46206853
-------------------------------------------------------------------------------
variable     |   dp/dx   Std. Err.    z     P>|z|  [    95% C.I.    ]       X
-------------+-----------------------------------------------------------------
p            |                                                                 
       beach |  .000609   .000061    9.97   0.000   .000489   .000729    103.42
     charter | -.006243   .000441  -14.15   0.000  -.007108  -.005378    84.379
        pier |  .000764   .000071   10.69   0.000   .000624   .000904    103.42
     private |   .00487   .000452   10.77   0.000   .003983   .005756    55.257
-------------------------------------------------------------------------------

Pr(choice = pier|1 selected) = .06584968
-------------------------------------------------------------------------------
variable     |   dp/dx   Std. Err.    z     P>|z|  [    95% C.I.    ]       X
-------------+-----------------------------------------------------------------
p            |                                                                 
       beach |  .000087   .000016    5.42   0.000   .000055   .000118    103.42
     charter |  .000764   .000071   10.69   0.000   .000624   .000904    84.379
        pier | -.001545   .000138  -11.16   0.000  -.001816  -.001274    103.42
     private |  .000694   .000066   10.58   0.000   .000565   .000822    55.257
-------------------------------------------------------------------------------

Pr(choice = private|1 selected) = .41959373
-------------------------------------------------------------------------------
variable     |   dp/dx   Std. Err.    z     P>|z|  [    95% C.I.    ]       X
-------------+-----------------------------------------------------------------
p            |                                                                 
       beach |  .000553   .000056    9.88   0.000   .000443   .000663    103.42
     charter |   .00487   .000452   10.77   0.000   .003983   .005756    84.379
        pier |  .000694   .000066   10.58   0.000   .000565   .000822    103.42
     private | -.006117   .000444  -13.77   0.000  -.006987  -.005246    55.257
-------------------------------------------------------------------------------

All the own-price elasticities are negative and the cross-price elasticities are positive which makes sense. The own-price elasticity of -.001249 means that a 1 Dollar increase from the mean of p (price) of fishing at the beach reduces the probability that beach fishing is chosen by 0.001249 for an individual with mean income and mean q (fish caught).
So all elasticities are expressed relative to the mean values of income, p, and q.

The cross-price elasticity .000609 tells you that if the price for fishing from a charter boat increases by 1 Dollar, the probability that beach fishing is chosen increases by .000609. The result is not in matrix format but you can easily take each dp/dx block from the results (each of which is a column for your final matrix) and put them together as a matrix of own- and cross-price elasticities where the own-price elasticities are on the principal diagonal and the cross-price elasticities are off diagonal like this:

             beach   charter      pier   private
------------------------------------------------
  beach | -.001249   .000609   .000087   .000553
charter |  .000609  -.006243   .000764    .00487
   pier |  .000087   .000764  -.001545   .000694
private |  .000553    .00487   .000694  -.006117

The fishing choice example seems trivial but it should now be straightforward for you to apply your own demand estimation problem to the code provided. For more information on estimating these price elasticities or the asclogit command have a look at Cameron and Trivedi (2009) "Microeconometrics Using Stata".

$\endgroup$
  • 2
    $\begingroup$ The reference is: Cameron and Trivedi (2009), Microeconometrics Using Stata. $\endgroup$ – Roberto Ferrer May 25 '14 at 22:33
  • $\begingroup$ Oh yes, you're right. Thanks for pointing out the typo. $\endgroup$ – Andy May 25 '14 at 22:43
  • $\begingroup$ @Andy thank you. It's very thorough and should help my understanding. $\endgroup$ – Black May 27 '14 at 13:35
1
$\begingroup$

If elasticity is the changes in probability as a result of 1% change in an independent variable, then first you have to:

1- Calculate probability of model, in stata predict, p1

2-Increase interested variable by 1%, in stata: var*1.01

3-Again calculate probability, predict, p2

4-The differences between two probabilities are elasticity. E=average(p2-p1)

$\endgroup$
0
$\begingroup$

The equations in above answer are the marginal effect, not the elasticity. Please check a seminal book by Prof. KE, Train.

See pp.57--60 in https://eml.berkeley.edu/books/choice2nd/Ch03_p34-75.pdf.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.