• No results found

Bank Failures and Skill Composition

In document Essays on Empirical Banking (Page 117-120)

3.8 Additional Robustness Tests

3.8.2 Bank Failures and Skill Composition

Under the Neyman-Rubin framework, one crucial assumption to establish a causal relationship is the stable unit treatment value assumption (SUTVA), where treatment assignment to one group does not affect the potential outcome for the other group (Rubin (1977)). Following the Neyman-Rubin framework, we would like to give more credibility to the magnitudes of our coefficients and test whether the results we pre- sented so far suffer from not satisfying SUTVA.

In our setting, if bank failures induce a change in skill composition and distri- bution in communities with no experience of bank failures, we may violate SUTVA. One can possibly think of a scenario of mobility of skilled workers towards unaffected PUMAs, driving down skilled wages and, hence, decreasing wage disparity relative to affected PUMAs. Although PUMA-year fixed effects fix all factors changing in community and year dimensions within a community (intra-community), it cannot take into account the inter-community effects. Similarly, there could be switching of skills between different sectors and occupation types.

One test of showing that SUTVA is met in several dimensions of our data (PUMA, sector and occupation types), looks at whether bank failures change the skill com- position and distribution. In particular, we concentrate on specifications with within estimation in which skill composition and distribution are regressed on bank failures. Finding significant asymmetries in skill composition and distribution within PUMAs, sectors and occupation types after bank failures would lead us to conclude that our

Table 3.14: Skill composition and distribution within PUMAs and sectors

Within PUMA Within sector

Skill composition Skill distribution

(1) (2) (3) (4) (5) (6) (7) (8)

EDUC Level of education EDUC EDUC EDUC Level of education Level of education Level of education

FAILED -0.000469 -0.00125 -0.000515 0.000175 0.0000308 -0.00000771 -0.00165 -0.00421

(0.000920) (0.00294) (0.000991) (0.00107) (0.00181) (0.00317) (0.00328) (0.00520)

Controls Yes Yes Yes Yes Yes Yes Yes Yes

Extra controls Yes Yes Yes Yes Yes Yes Yes Yes

PUMA FE Yes Yes Yes No Yes Yes No No

Year FE Yes Yes Yes No Yes Yes No No

Ind FE No No Yes No No Yes No No

PUMA-Ind FE No No No Yes No No Yes No

Ind-Year FE No No No Yes No No Yes No

PUMA-Ind-Age FE No No No No Yes No No Yes

Ind-Age-Year FE No No No Yes Yes No No Yes

N 1451995 1451995 1105094 1093746 802483 1105094 1093746 802483

R2 0.841 0.641 0.828 0.846 0.937 0.653 0.692 0.877

Notes: The sample period is 2008-2010. EDUC is a dummy variable taking the value of one (1) for individuals with an undergraduate degree or above and zero (0) otherwise. Level of education is categorical variable based on years of schooling. FAILED is a dummy variable taking the value of one (1) for individuals residing in PUMAs that experience at least one bank failure during the sample period and zero (0) otherwise. KNOWDEP and R&D are sector-level time-invariant variables. Each column includes one- or multi-way fixed effects. Controls include the following variables: Experience, Experience2Foreign born, Race, Female, EDUC*Experience and EDUC*Experience2. Extra controls include the following variables: Age, Age2, Married, Married*Child, Female*Child. Whenever Age fixed effects are included, Age and Age2 are dropped from the regressions. Standard errors are shown in parentheses and are clustered by PUMAs. *, **, *** indicate significance at the 10%, 5% and 1% levels.

results are contaminated by not satisfying SUTVA. Table 3.14 presents the estima- tion results for within PUMA and within sector regressions. In Table 3.14 the first two columns refer to within PUMA estimates and the remaining six columns refer to within sector fixed effects. EDUC is, as defined before, to be skilled or not (cor- responding to skill composition) and “Level of education” is finer ordered categories of schooling such as elementary, high school, college and so on (corresponding to skill distribution). The estimation results clearly show that neither the skill composi- tion nor the distribution changes within sectors and PUMAs. In other words, there is no significant switching of one specific type of worker from one PUMA or sector to the other. This suggests that our baseline results can unlikely be explained by not satisfying SUTVA across PUMA and sector dimensions.

Table 3.15: Skill composition and distribution within occupation types

Skill composition Skill distribution

(1) (2) (3) (4) (5) (6)

EDUC EDUC EDUC Level of education Level of education Level of education FAILED -0.000666 -0.000317 -0.000160 -0.000738 -0.000221 -0.000757

(0.000998) (0.00110) (0.00234) (0.00284) (0.00319) (0.00589)

Controls Yes Yes Yes Yes Yes Yes

Extra controls Yes Yes Yes Yes Yes Yes

PUMA FE Yes No Yes Yes No No

Year FE Yes No Yes Yes No No

Occp FE Yes No No Yes No No

PUMA-Occp FE No Yes No No Yes No

Occp-Year FE No Yes No No Yes No

PUMA-Occp-Age FE No No Yes No No Yes

Occp-Age-Year FE No Yes Yes No No Yes

N 1105094 1077346 729649 1105094 1077346 729649

R2 0.833 0.858 0.957 0.718 0.766 0.937

Notes: The sample period is 2008-2010. EDUC is a dummy variable taking the value of one (1) for individuals with an undergraduate degree or above and zero (0) otherwise. Level of education is a categorical variable based on years of schooling. FAILED is a dummy variable taking the value of one (1) for individuals residing in PUMAs that experience at least one bank failure during the sample period and zero (0) otherwise. KNOWDEP and R&D are sector-level time-invariant variables. Each column includes one- or multi-way fixed effects. Controls include the following variables: Experience, Experience2Foreign born, Race, Female, EDUC*Experience and EDUC*Experience2. Extra controls include the following variables: Age, Age2, Married, Married*Child, Female*Child. Whenever Age fixed effects are included, Age and Age2 are dropped from the regressions. Standard errors are shown in parentheses and are clustered by PUMAs. *, **, *** indicate significance at the 10%, 5% and 1% levels.

ican Community Survey provides information on occupation types consisting of 539 specific occupational categories arranged into 23 high-level occupational groups. This classification was created on the basis of the Standard Occupational Classification (SOC) Manual: 2010, published by the Executive Office of the President, Office of Management and Budget (OMB) (US Census Bureau (2017)). Table 3.15 displays the results of within occupation type regressions.

Table 3.15 reiterates the findings that there is no evidence for significant asymme- tries in skill composition and occupation as a result of bank failures. Overall, Table 3.14 and Table 3.15 indicate that our baseline results can unlikely be explained by not satisfying SUTVA across PUMA, sector and occupation types.

3.8.3

Migration, Occupation Types, Stable Sample, Winsorizing data

In document Essays on Empirical Banking (Page 117-120)