Estimating the modified Jones model by industry and year using panel data

[Kiki Qu]

Hi all. I'm studying the relationship between restatements and firms' earnings management behaviour. I'm using DA to capture such behaviour. The sample period is 2010-2019. Let's say a firm restated in 2015, I'm interested in its DA in period 13, 14, 16, 17 (two years before and two years after). Data in t-3 (2012 in this case) and t (2015) is also included as this is a panel data. I have all the data transformed into the delta form, it is just the last step that I can never get right. I attached a file in which all the data are processed and are ready for the last step.

below is the code I first used:

gen Jones=.
forval y = 2010 (1) 2019{
forval i = 1 (1) 136{
display `i'
display `y'
reg scaled_TACC inverse_lagTA term2 scaled_PPE if `i' == Industryid&`y'== YEAR, noconstant
predict r if `i'== Industryid&`y'== YEAR, resid
replace Jones=r if `i'== Industryid&`y'== YEAR
drop r
}
}

But this gives me nothing and STATA keeps showing r451. Then I reformed it and added a line:

​​​​​​ gen Jones=.
forval y = 2010(1)2019{
   forval i = 1(1) 136{
      display `i'
      display `y'
      capture noisily{
         reg scaled_TACC inverse_lagTA term2 scaled_PPE if `i'== Industryid & `y'== YEAR, noconstant
         predict r if `i'== Industryid & `y'== YEAR, resid
         replace Jones_2005_TAC2=r if `i'== Industryid & `y'== YEAR 
         drop r
      }
   }
}

Then the code starts running but, still, I have quite a lot of missing values in my result. Really hope someone could help me. Thanks in advance!!

[William Lisowski]

Your dataset has just 2125 observations, spread over 136 Industries and 10 years: 1360 different combinations you hope to to run. The table below shows that of those 1360 combinations, fewer than 5% of them have 5 or more observations for which none of the variables in the model are missing.

       nobs |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |        494       36.32       36.32
          1 |        517       38.01       74.34
          2 |        193       14.19       88.53
          3 |         66        4.85       93.38
          4 |         28        2.06       95.44
          5 |         10        0.74       96.18
          6 |         16        1.18       97.35
          7 |          6        0.44       97.79
          8 |          4        0.29       98.09
          9 |         10        0.74       98.82
         10 |          4        0.29       99.12
         11 |          3        0.22       99.34
         12 |          1        0.07       99.41
         13 |          1        0.07       99.49
         15 |          1        0.07       99.56
         16 |          1        0.07       99.63
         18 |          1        0.07       99.71
         23 |          1        0.07       99.78
         24 |          1        0.07       99.85
         29 |          1        0.07       99.93
         32 |          1        0.07      100.00
------------+-----------------------------------
      Total |      1,360      100.00

I am afraid I know nothing of the substance of the Jones model. In an earlier discussion I participated in at

https://www.statalist.org/forums/for...ing-panel-data

the member wrote that the recommended number of observations is 40. You have no combination comes even close to that threshold.

In summary, the problem is not a Stata problem. Either your data is insufficient to support the analysis you want to do, for most combinations of industry and year, or your understanding of the Jones model is incomplete.

[Kiki Qu]

Thank you, William! I used SIC code to classify industry. Maybe I should try using FF industry classification code, in which circumstance it might work!

[Kiki Qu]

I found this link that tells you how to match SIC code with FF industry code, but I'm not sure how to code in my case. The link is attached. Could you please give me some direction? Really appreciate!!https://volkovanotes.wordpress.com/2...ch-industries/