I advice you used ratios for your financial performance variables that way you do not have to deal with the issue of transformation. For profitability you may use ROE or ROA while working capital you may use Current ratio or Acid Test Ratio.

Ok Thank you for your constructive answer Ismaila Yusuf

regressions usually capture variations among observations and hence adding the same value to all observations can help you remove the negative value and comfortably transform your data. For instance, suppose you have A, B , and C with values 20, -15, and - 5 respectively, adding 30 to each of them gives you a new value of 50, 15, and 25 respectively. The R2, F-stat and coefficient estimates remain the same, but the constant term would be overstated by the amount you added. Therefore, you should adjust it by deducting from it the value you added. For example if the estimated value of the constant is 25, the adjusted value would be -5( i.e 25-30). In addition, you should use the original values in your descriptive statistics. 

Oh Really great answer. Thanks dear Ashenafi Fanta

Hi pradip, I am kind of doing a similar work with ownership. Do you care to share your work? My email id would be [email protected].  Thanks.

Hello everyone,

Fantastic insight from Fanta :) Never seen things like that.

In the similar context, I have a question for my regression Analysis too. I am taking average of last 3 years D/E ratio and taking it in natural log form. Excel is generating values in parenthesis () in some context. I need to ascertain the interpretation of such values and how will they be interpreted once in I use them in regression as Independent variable. Any help is much appreciated.

Another question that I have is I am averaging accounting variables for past three years before using them in regression as Independent variables and in plenty of cases the values are missing for all the three years. How do we move ahead in such situations?

Thank you very much!

Dear Vineet,

If you have a three year data, you should rather use panel data models instead of averaging. Averaging(a.k.a temporal aggregation) leads to loss of the time series information, limiting your analysis to cross sectional comparisons. For instance, if you have D/E ratio of 0.6, 0.4, and 0.2 for year 1,2 and 3 respectively, your average D/E for the firm would be 0.4 and doing so leads to loss of the information that the firm's leverage has been continuously declining over the three year period. If you use panel data models, you will have both dimensions of the data and it will improve your results.

As regards, missing values you have two options: (1) eliminate cases with missing values and this will reduce the number of observations (2) impute missing values after ensuring that they occur randomly. You can perform multiple imputation on IBM SPSS by going to Analysis>multiple imputation>Impute missing data value. IBM SPSS estimates missing values by automatically selecting the model. Once you have the imputed values, you can proceed with estimation of your model.

I hope that helps.

Ashenafi B Fanta

Thank you very much Ashenafi. I realized loss of time series information when I aggregated my data.

Now my work needs me to run Regression with some aggregate accounting variables  as independent variable - like average D/E, Average NP margin, Average ROA etc. since it is a published work, I would assume that they just need cross-sectional analysis. But I agree it would be nice if I can improve upon that and some how capture Time series as well. I am not sure how to do that.

Besides, following your suggestions to Pradiptaarthi, I tried  running regression by adding constants to my X Variables. Indeed, I get same coefficients but I failed to get the difference in constant values. I just used random data for X, and Y to try it out. Have I missed out something?

Thanks again,

Vineet