1. General context
Variation in Federal Reserve Bank policies adopted in 1920
National banks in my data set
2. Focus on split states
The cleanest possible identification of treatment effects comes from the split states in my sample (Indiana, Kentucky, New Jersey). In the split state setting, potential policy discontinuities across state borders can be ruled out as spurious drivers of treatment effects in my geographic regression discontinuity framework.
a) Identification strategy
Interbank links of treated national banks in split states
One crucial aspect of my identification strategy is that treated national banks do not have interbank correspondents located in the non-treated territory of the split state. The existence of regulatory arbitrage within the split state (across the Federal Reserve district border) is particularly problematic because it could lead to a violation of the stable unit treatment value assumption (SUTVA): by blowing up the balance sheet of control units, this arbitrage channel could induce an upward bias in the estimated treatment effect.
In the maps below, I show the interbank links of national banks located in the treated territory of Kentucky and New Jersey (each red dot represents a bank location). Kentucky is split between the eighth Federal Reserve district which used the microprudential tool and the fourth Federal Reserve district which did not change policy stance. New Jersey is split between the second Federal Reserve district which leant against the wind and the third Federal Reserve district which did not change policy stance.
Interbank links across Federal Reserve district borders but within state borders do practically not exist. In the case of Kentucky, not a single treated bank entertained a correspondent relationship with a bank in the control territory of the split state. In the case of New Jersey, only one treated bank was linked to a New Jersey control bank (in Trenton). However, by January 1921, even this relationship had disappeared.
Kentucky banks located in district 8 (Jan 1920)
New Jersey banks located in district 2 (Jan 1920)
b) Mapping outcome variables
Prudential policy & leverage ratio: Kentucky (Jan 1920 – Jan 1921)
The map below shows changes in bank-level leverage ratios between January 1920 and January 1921 west and east of the Federal Reserve district border (vertical dashed line). West of the border, banks were subject to the prudential policy. The short horizontal lines in the diagrams represent the mean change for entire state of Kentucky. The raw data show that bank-level leverage evolved differently west and east of the Federal Reserve district border splitting Kentucky. On average, untreated banks in the East increased leverage (much) more than treated banks in the West (many of which even substantially reduced leverage). The size of the bar charts in the map below is proportional to pre-treatment leverage ratios as measured in January 1920. The color of the charts is chosen according to the percentile range in which a given bank’s change in the leverage ratio is located: red 0-20, orange 21-40, yellow 41-60, green 61-80, dark green 81-100.