Log conversion problem

[thefan]

I have this question....
in order to linearize a value one can convert a value into log form. but what if the value is negative...suppose something like trade balance (as export - import = Tb and when imports exceed export we get a negative value). In such a case how does one linearize TB? is it statistically correct to ignore the (-) sign and only convert the absolute value.
for Eg: if TB = -2343
log TB= ignore the (-) sign and log 2343 =ln 7.759187439 and later we include the (-) sign so ln -7.759187439......is this correct?

or how does one convert negative values into ln values?

[startz]

I have this question....
in order to linearize a value one can convert a value into log form. but what if the value is negative...suppose something like trade balance (as export - import = Tb and when imports exceed export we get a negative value). In such a case how does one linearize TB? is it statistically correct to ignore the (-) sign and only convert the absolute value.
for Eg: if TB = -2343
log TB= ignore the (-) sign and log 2343 =ln 7.759187439 and later we include the (-) sign so ln -7.759187439......is this correct?

or how does one convert negative values into ln values?

You can't ignore the sign. If the values are very negative, then using logs really doesn't make sense. If they are only slightly negative, then one trick is to add a constant to all observations before taking logs.

A procedure which may be safer is to use ratios rather than logs, for example TB/GDP.

[thefan]

what about an index? you know by taking an year as the base...like the following

year TB
1999 2343
2000 2567
2001 3091

so taking 2000 as the base year 2000=100

year TBindex
1999 91.27
2000 100
2001 120.41 where the index was constructed as follows suppose its for 1999


1999 TB / 2000 TB * 100 etc.


is this way statistically correct?

can one convert all variables (such as GDP, Exchange rate) in a model into indeces and run an OLS ?