Instrument for banksize

[Hans de Laak]

Dear everyone,

I am currently investigating the relationship between bank size and banking risk. This relationship is potentially endogenous, so I'd like to test whether an instrument is necessary in order to analyze the relationship. However, after long research, I have still not been able to find a valid instrument for bank size. I am measuring bank size in (log) assets, and risk is measured using a factor analysis on 20 determinants of risk. If anyone has tips, or help to help mitigate this endogeneity problem, it would be greatly appreciated!

Thanks

[Carlo Lazzaro]

Hans:
welcome to this forum.
May the number of bank branches/subsidiaries/tellers be resonable as an instrument ?(disclaimer: temptative answer).

[Hans de Laak]

Dear Carlo,

Thanks for your response. I have data on the number of managers per bank, I was already thinking in this way, but I am unsure if this is a valid instrument, since there potentially might be a relationship between managers and risk, other than through size. For example (just brainstorming here, do not know whether it has been proven), fewer managers might mean they have less superiors to report to and thus "easier" to take on more risk. I might be overthinking this, but this is what I remember from my econometrics class. For the other parameters I do not have data ready at my disposal, I could potentially hand-collect but this is possibly too time consuming for my time frame. (Sample is over 800 banks)

[Hans de Laak]

What do you guys think of number of years of financial reporting available as instrument for size?

[Carlo Lazzaro]

Hans:
unfortunately I do not know the literature of your research field, but the lower the number of managers, the higher the potential risk taken sounds reasonable.
That said, aren't banks compelled by law and central bank to provide yearly financial reports?

[Hans de Laak]

Dear Carlo,

Yes they are. My 'instinct' was that the larger the number of years they have been reporting, the older the bank, and possibly the larger the bank, as you would say that 'older' banks tend to be larger. However, I am not sure this would work.

[Carlo Lazzaro]

Hans:
in each scientific field, "good" instruments are often related to specific acceptability standards
Hopefully you can get more substantive details with skimming through the literature of your research field and/or discussing this issue with of your colleagues.

[Hans de Laak]

Dear Carlo,

Yes, I will continue my research, and eventually I am hopeful I will find the "right" instrument.

[Phil Bromiley]

To add to Carlo's helpful comments, depending on how many periods you have of serial correlation, lagged values of the right-hand side variables can be legitimate instruments as long as they are lagged more than the serial correlation.

Another way to do this would be to simply lag all your right hand side variables a year or two and argue that it is unlikely that risk in T is is influencing bank size in T-1 or T-2.

[Hans de Laak]

Thanks Phil, I will try to use this as well.