Cox regression application in business/finance

Hi all,

I hope this is the right forum to ask as cox regression is usually use by bio-statisticians.

My research is about companies which producing either product A or product B. This companies will either distress (1) or stay healthy (0). The time of production for product A/B are varies, but i set the end time of observation as the same, i.e. year 2013. The data is in annual form.

I am using SPSS and following are the variables:

Dependent variable: distress (1), healthy (0) [selected as status in SPSS]

Time to event : years of after production until distress/year 2013 [selected as time in SPSS]

Time dependent: T*_SIZE

Independent variables/covariates: 5 variables including SIZE. All are in the form of financial ratios

My question is:
1. whether the data is correct
2. can i have more than one time dependent variable? My covariates are all financial variables and it will change over time, hence time dependent?
3. since the covariates are changing over time, should I take average data for all observation period from production year until year of distress/year 2013?

Appreciate any feedback. I thank you in advance !


Fortran must die
I spent a lot of time studying Cox, and related methods this year, but am certainly no expert.

1) I am not sure what you mean by correct. I don't see a variable that tells the software whether someone is censored or not which is normally required.
2) I don't think there is any limit to the number of time varying independent variables. If the variables change over time you have to use a special form of Cox (technically I don't think it is Cox, but it is in the Cox family of methods). In SAS you have to handle the data in a special way I think for these time varying covariates - you should check how SPSS handles this.
3) There are specific ways of addressing this issue, but I have not seen that approach.

This discusses this issue although it is with R code.

You may already have seen this, but I include it if not. It is SPSS's introduction to this issue.

Unfortunately I know too little about SPSS to comment beyond that point.
Hi Noetsi,

Thanks for your reply. Yes, me too, spending a lot of time to understand Cox.

1. What I mean by correct is whether the data is correctly grouped into dependent/independent/time dependent variables. In my case, the companies are censored if they
discontinue the production of product A/B. For that I categorized the company as (0).

I read about the SPSS intro to Cox, but it doesnt help me to apply Cox in my case. Please let me know if you have any links/suggestion of books etc for application of Cox in business/finance.

Also, since the companies are either producing product A or B, should I run the analysis separately? In other words, should i have one set of data for companies only producing product A, and the other set of data for companies only producing product B?

Thank you.


Fortran must die
Although I have not seen this addressed I think you should run these analysis separately. In part because I have not seen the analysis run the way you are suggesting, but also there could be signficant differences in the process of producing A and B. I don't see the logic of combining what is essentially two entirely different samples this way.

Most of the survival analysis I have seen related to medical research. I have not seen it applied to business or finance, but I don't think that matters. As long as you meet the assumptions of the method and have a valid use for it, the fact that others have not used it yet is besides the point.

I know a good book for Cox (Paul Allison's treatement of it) but it is for SAS not SPSS. It does not deal with business - again I have not seen any Cox models applied to business.
yes, I didnt address the type of the product in the sample, for that I am asking whether the data is correct. Besides, I do not know how SPSS handling this variables ( of product A/B). The products are financial products and it have some similarities. I am looking forward to run the same analysis in Stata. Thanks !