Dear all,
I am writing my masters thesis on the effect of market concentration on risk-taking behavior of banks in the U.S.
My three risk-taking variables are loan-loss provisions, the Z-Score and uninsured deposits. I calculate the main dependent variable, the HHI, using originated loans by each bank instead of assets to calculate market share, hence I am proxying market share (compared to the traditional way of calculating it).
Approach 1: I created quartiles based on the following metrics: total assets, liquidity (Cash/Total Assets ratio) and deposit funding (Total Deposits/Total Assets ratio). I do this so that I can restrict my regressions based on the bottom and top 25% of each respective category. Hence, I would have 6 regressions per each risk variable, totalling 18 total regressions. I imagine this would create an isolated effect of not market concentration but also for all other control variables that I am using. This way I can more clearly compare the effects of a bank that finds itself in the bottom versus the top quartile of assets.
Approach 2: Alternatively, I can also create interaction terms that can interact the HHI with a respective top or bottom quartile to indicate the effect of a bank being in the respective category, hence I would have 6 interaction terms between the HHI with the bottom and top 25% of assets, liquid and deposit funding. This approach will include the full interaction terms in an unrestricted sample regression so I would have only 1 main regression per each risk-taking variable. This approach allows me to compare the effect of the quartiles while maintaining the full sample.
I am struggling however to address the advantages and disadvantages to both approaches as I have already conducted my research based on the first approach but I am having doubts regarding if is correct to move forward with the results.
Thank you,
Viktor
I am writing my masters thesis on the effect of market concentration on risk-taking behavior of banks in the U.S.
My three risk-taking variables are loan-loss provisions, the Z-Score and uninsured deposits. I calculate the main dependent variable, the HHI, using originated loans by each bank instead of assets to calculate market share, hence I am proxying market share (compared to the traditional way of calculating it).
Approach 1: I created quartiles based on the following metrics: total assets, liquidity (Cash/Total Assets ratio) and deposit funding (Total Deposits/Total Assets ratio). I do this so that I can restrict my regressions based on the bottom and top 25% of each respective category. Hence, I would have 6 regressions per each risk variable, totalling 18 total regressions. I imagine this would create an isolated effect of not market concentration but also for all other control variables that I am using. This way I can more clearly compare the effects of a bank that finds itself in the bottom versus the top quartile of assets.
Approach 2: Alternatively, I can also create interaction terms that can interact the HHI with a respective top or bottom quartile to indicate the effect of a bank being in the respective category, hence I would have 6 interaction terms between the HHI with the bottom and top 25% of assets, liquid and deposit funding. This approach will include the full interaction terms in an unrestricted sample regression so I would have only 1 main regression per each risk-taking variable. This approach allows me to compare the effect of the quartiles while maintaining the full sample.
I am struggling however to address the advantages and disadvantages to both approaches as I have already conducted my research based on the first approach but I am having doubts regarding if is correct to move forward with the results.
Thank you,
Viktor
