Maximum Likelihood estimation

[Alex Izquierdo]

Hi everyone,

For my master's thesis I'm trying to do a maximum likelihood estimation. After working with my tutor, I've developed the do file, which I've saved with the name of Code:

 use "DataML2.dta"
sort id
 program drop _all
 program define MLSchool
 version 14.0
 args todo b lnf
 tempvar val lnsigma lnrho //where val = b*X_ij
 mleval `val' = `b', eq(1)
 mleval `lnsigma' = `b', eq(2)
 qui {
 replace sigma = exp(`lnsigma')
 by id : egen val1 = sum((school ==choice1 )*`val')
 by id : egen val2 = sum((school ==choice2 )*`val')
 by id : egen val3 = sum((school ==choice3 )*`val')
 by id : egen val4 = sum((school ==choice4 )*`val')
 by id : egen val5 = sum((school ==choice5 )*`val')
 by id : egen val6 = sum((school ==choice6 )*`val')
 by id : egen val7 = sum((school ==choice7 )*`val')
 by id : egen sigma1 = sum((school ==choice1 )*sigma)
 by id : egen sigma2 = sum((school ==choice2 )*sigma)
 by id : egen sigma3 = sum((school ==choice3 )*sigma)
 by id : egen sigma4 = sum((school ==choice4 )*sigma)
 by id : egen sigma5 = sum((school ==choice5 )*sigma)
 by id : egen sigma6 = sum((school ==choice6 )*sigma)
 by id : egen sigma7 = sum((school ==choice7 )*sigma)
 
 
 
 * Simulated log lihelihood
 
 forvalues t=1/10 {
 
 *no variable repetition

 
 *generate noises and total values
 replace epsilon = sigma*norm`t'
 replace totval= `val'+epsilon   //+outside*phi
 by id : egen totalval0 = 0

 

 
 *likelihood of NOT
 by id: replace likenot = sum(ignore* ln(normal((totalval0-`val')/sigma)))
 by id: replace likenot = exp(likenot[_N])
 
 by id: replace likenotn = sum(ignoren*ln(normal((totalval0-`val')/sigma)))
 by id: replace likenotn = exp(likenotn[_N])



* Value and other things for 7th (last) option, naive and sophisticated
drop crit7
 by id: egen crit7= max(prob7*(totval-totalval0)-((position<8)+ignore)*10000000000)
 replace like7 = normal(-(crit3/q3corr+totalval0-val7)/sigma7)
 replace e7= -sigma7*invnormal(u7`t'*like7)
 
 drop crit7n
 by id: egen crit7n=max(totval-((position<8)+ignore)*10000000000)
 replace like7n = normal(-(crit7n-val7)/sigma7)
 replace e7naive = -sigma7*invnormal(u7`t'*like7n)
 replace totval7=val7+e7
 replace totval7=totalval0 if roundnot<8
 replace res7=q7*totval7+(1-q7)*totalval0
 replace totval=totval7 if choice7==school
  
 replace totval7n=val7+e7naive
 replace totval7n=max(crit7n,totalval0) if roundnotn<8
 
* Value and other things for 6th option

 drop crit6
 by id: egen crit6= max(prob6*(totval-res6)-((position<7)+ignore)*10000000000)
 replace like6 = normal(-(crit6/q6corr+res7-val6)/sigma6)
 replace e6= -sigma6*invnormal(u6`t'*like6)
 
 replace like6n = normal(-(totval7n-val6)/sigma6)
 replace e6naive= -sigma6*invnormal(u6`t'*like6n)
 
 replace totval6=val6+e6
 replace totval6=totalval0 if roundnot==6
 replace res6=q6*totval6+(1-q6)*res7
 replace totval=totval6 if choice6==school
 
 replace totval6n=val6+e6naive
 replace totval6n=max(crit7n,totalval0) if roundnotn==6
 
* Value and other things for 5th option

 drop crit5
 by id: egen crit5= max(prob5*(totval-res6)-((position<6)+ignore)*10000000000)
 replace like5 = normal(-(crit5/q5corr+res6-val5)/sigma5)
 replace e5= -sigma5*invnormal(u5`t'*like5)
 
 replace like5n = normal(-(totval6n-val5)/sigma5)
 replace e5naive= -sigma5*invnormal(u5`t'*like5n)
 
 replace totval5=val5+e5
 replace totval5=totalval0 if roundnot==5
 replace res5=q5*totval5+(1-q5)*res6
 replace totval=totval5 if choice5==school
 
 replace totval5n=val5+e5naive
 replace totval5n=max(crit6n,totalval0) if roundnotn==5

* Value and other things for 4th option
 drop crit4
 by id: egen crit4= max(prob4*(totval-res5)-((position<5)+ignore)*10000000000)
 replace like4 = normal(-(crit4/q4corr+res5-val4)/sigma4)
 replace e4= -sigma4*invnormal(u4`t'*like4)
 
 replace like4n = normal(-(totval5n-val4)/sigma4)
 replace e4naive= -sigma4*invnormal(u4`t'*like4n)
 
 replace totval4=val4+e4
 replace totval4=totalval0 if roundnot==4
 replace res4=q4*totval4+(1-q4)*res5
 replace totval=totval4 if choice4==school
 
 replace totval4n=val4+e4naive
 replace totval4n=max(crit5n,totalval0) if roundnotn==4
 
* Value and other things for 3rd option

 drop crit3
 by id: egen crit3= max(prob3*(totval-res4)-((position<4)+ignore)*10000000000)
 replace like3 = normal(-(crit3/q3corr+res4-val3)/sigma3)
 replace e3= -sigma3*invnormal(u3`t'*like3)
 
 replace like3n = normal(-(totval4n-val3)/sigma3)
 replace e3naive= -sigma3*invnormal(u3`t'*like3n)
 
 replace totval3=val3+e3
 replace totval3=totalval0 if roundnot==3
 replace res3=q3*totval3+(1-q3)*res4
 replace totval=totval3 if choice3==school
 
 replace totval3n=val3+e3naive
 replace totval3n=max(crit4n,totalval0) if roundnotn==3
  
  
* Value and other things for 2nd option

 drop crit2
 by id: egen crit2= max(prob2*(totval-res3)-((position<3)+ignore)*10000000000)
 replace like2 = normal(-(crit2/q2corr+res3-val2)/sigma2)
 replace e2= -sigma2*invnormal(u2`t'*like2)
 
 replace like2n = normal(-(totval3n-val2)/sigma2)
 replace e2naive= -sigma2*invnormal(u2`t'*like2n)
 
 replace totval2=val2+e2
 replace totval2=totalval0 if roundnot==2
 replace res2=q2*totval2+(1-q2)*res3
 replace totval=totval2 if choice2==school
 
 replace totval2n=val2+e2naive
 replace totval2n=max(crit3n,totalval0) if roundnotn==2
 
 
 * Value and other things for 1st option
 drop crit1
 by id: egen crit1= max(prob1*(totval-res2)-((position<2)+ignore)*10000000000)
 replace like1= normal(-(res2-val1+crit1/q1corr)/sigma1)
 replace like1n= normal(-(totval2n-val1)/sigma1)
 
 
 *total flow likelihood
 
 replace flowrat = like1*like2^(roundnot>2)* ///
 like3^(roundnot>3)*likenot^(roundnot<4)
 replace flownai = like1n*like2n^(roundnotn>2)* ///
 like3n^(roundnotn>3)*likenotn^(roundnotn<4)
 
 replace cummrat = cummrat+flowrat/10
 replace cummnai = cummnai+flownai/10
 
}
 
  
  replace cumm = max(cummrat,cummnai)
  replace rat = (cummrat>=cummnai)
  replace cumm=1 if last ==0 | cumm>1
  replace cumm=0.0000000000000000001 if cumm<0.0000000000000000001
  mlsum `lnf'= ln(cumm) if last==1
  *drop val1 val2 val3 sigma1 sigma2 sigma3
  replace cummrat = 0
  replace cummnai = 0
  }
 end

Afterwards, I type the following in Stata command:

preserve
do Code
ml model d0 MLSchool (poverty monoparent sqdist dist1000 dist5000 change) /lnsigma
(until here everything works fine).

Finally, before starting the estimation, I type:
ml check

And I obtain the following error:

Test 1: Calling MLSchool to check if it computes log likelihood and
does not alter coefficient vector...
FAILED; MLSchool returned error 133.

And investigating further, I find (between a ton more of text):

---------------------------------------------------- end findfile ---
- if (`"`r(fn)'"' == "") {
= if (`""' == "") {
- di as error "unknown egen function `fcn'()"
= di as error "unknown egen function 0()"
unknown egen function 0()
- exit 133
}
}
---------------------------------------------------------- end egen ---

It also appears some error 198, but since this is the one highlighted I imagine this is the problematic one.
I've been trying to find a solution for it, but I haven't been able to. I would very much appreciate it if someone had any suggestions on what I should do.
Thank you very much for your help.

Best,

Àlex

[Rich Goldstein]

I have not gone thru your code in detail, but a quick search on -egen- found the following problem: "
by id : egen totalval0 = 0 + epsilon " you have not included a function there and that is the reason for the message; looking at just that line, it appears that you want -gen- rather than -egen

edited to fix typo

[Alex Izquierdo]

That seems to solve the problem for now. There was a function originally in that line and I forgot to change it. Thank you very much.

[Ayub UOM]

Hello experts
I want to calculate beta coeffiecnt using maximum likelihood estimation and rolling ten-year windows for firm-year wise,but i dont know how to calculate it,could you please guide me how can i calculate it?
for explaintion i am giving the detail below

Following previous research (e.g., Lev 1983; Ali and Zarowin 1992) we measure earnings
persistence as the slope coefficient estimate,
Beta i,j, from an autoregressive model of order one (ARi) for annuals plit-adjustede arningsp er share (Xj,t) measured as firm j 's net income before extraordinary items in year t divided by the weighted average numberof outstanding shares during year t):
XJ,t =constant +Beta1,jt Xj,t-1 + vj,t. (2)
For each firm-year wise estimate Equation( 2) using maximum likelihood estimation and rolling ten-year windows. This procedure yields firm- and year-specific estimates of Beta1j
which capture the persistence of earnings.
.

.
looking forward for your guidence.
thanks