Does Disaster Risk Relate to Banks’ Loan Loss Provisions?

Figures & data

Figure 1. Disaster risk (DR) across counties in continental USA.

Note: Figure 1 presents a map chart of average disaster risk (DR) across counties in continental USA.

Table 1. Validation of disaster risk measure through the UN Sendai framework for disaster risk reduction.

Panel A: Descriptive statistics
Variable	N	Mean	Std. Dev.	Q1	Median	Q3
DRt	218,952	4.653	2.838	3	4	6
NHIt	218,952	12.033	2.220	10	12	14
POPt	218,952	10.263	1.460	9.315	10.150	11.094
EMPL%t	218,952	0.028	0.076	0.003	0.007	0.018
SOVIt	218,952	49.999	28.571	25.4	49.9	74.7
PTY_DAMAGEt	218,952	5.314	2.161	4.192	5.453	6.676
Panel B: Validation of disaster risk measure
	Column (1)	Column (2)	Column (3)
	DRt	DRt	DRt
Constant	−2.6512***	0.3912***	−1.1091***
	[−43.28]	[6.69]	[−19.66]
NHIt	0.0359***	0.0976***	0.0921***
	[12.49]	[35.04]	[34.61]
POPt	0.3610***	0.2117***	0.2728***
	[64.78]	[42.16]	[56.38]
EMPL%t	0.3286***	0.8422***	0.9625***
	[3.65]	[9.30]	[10.80]
SOVIt	0.0012***	0.0006***	0.0007***
	[6.15]	[3.60]	[4.60]
PTY_DAMAGEt	0.5826***	0.1621***	0.3371***
	[175.48]	[64.53]	[120.48]
Fixed Effects	Year-Quarter	State	Year-Quarter & State
Observations	218,952	218,952	218,952
R^2	0.1612	0.4667	0.5225

Note: This table reports the results for the validation approach our disaster risk measure (DR_t) through the UN Sendai framework for disaster risk reduction. Panel A reports summary statistics for disaster risk and disaster risk determinants. N is the number of county-quarter observations for which disaster risk and disaster risk determinants are available. Panel B reports estimation results for the following model validating our disaster risk measure (DR) through the UN Sendai framework for disaster risk reduction: $D{R}_t=f(\mathrm{HAZAR}{D}_t,\mathrm{EXPOSUR}{E}_t,\mathrm{VULNERABILIT}{Y}_t)$

Natural Hazard Index (NHI_t) is the natural hazard index is a multiple hazard measure that varies from 0 to 33. The higher the NHI_t, the higher the natural hazard in a county (National Center for Disaster Preparedness, NCDP, at Columbia University); Population (POP_t) is the natural logarithm of 1 plus the yearly number of individuals (both civilian and military) who reside in a given county (Bureau of Economic Analysis); Total employment (EMPL%_t) is the natural logarithm of 1 plus yearly total job at county level divided by the total job at country level (Bureau of Economic Analysis); Social Vulnerability Index (SoVI_t) is the social vulnerability index is a measure of social vulnerability to environmental hazards that varies from 0 to 100. The higher the SoVI_t, the more vulnerable a county (Hazard & Vulnerability Research Institute, HVRI, at University of South Carolina); Property damage (PTY_DAMAGE_t) is the natural logarithm of 1 plus property losses per capita from natural disasters over a 15-year span in each county-quarter (Spatial Hazard Events and Losses Database, SHELDUS). *, **, and *** indicate significance at the 0.10, 0.05, and 0.01 levels (two-tailed), respectively. t-statistics are presented in parentheses and are based on robust standard errors.

Table 2. Descriptive statistics.

Variable	Obs.	Mean	Std. Dev.	Q1	Median	Q3
LLPt	445,924	0.0011	0.0024	0.0000	0.0004	0.0011
DRt	445,924	4.9660	2.8456	3.0000	5.0000	7.0000
ΔNPAt+1	445,924	0.0001	0.0068	−0.0012	0.0000	0.0008
ΔNPAt	445,924	0.0001	0.0067	−0.0012	0.0000	0.0008
ΔNPAt-1	445,924	0.0001	0.0067	−0.0011	0.0000	0.0008
ΔNPAt-2	445,924	0.0001	0.0066	−0.0011	0.0000	0.0008
ΔLOANSt	445,924	0.0164	0.0532	−0.0122	0.0115	0.0384
EBTLLPt	445,924	0.0049	0.0041	0.0028	0.0045	0.0065
COt-1	445,924	0.0009	0.0018	0.0000	0.0003	0.0010
TIER1t-1	445,924	0.1075	0.0346	0.0854	0.0988	0.1192
ALLOWANCEt-1	445,924	0.0156	0.0082	0.0108	0.0136	0.0180
SIZEt-1	445,924	11.8729	1.1632	11.0903	11.7947	12.5564
BRANCHDIVt	445,924	23.2934	27.9751	0.0000	0.0000	50.0000
LoanTypest (residential)	445,924	0.7064	0.1975	0.5916	0.7411	0.8533
LoanTypest (commercial)	445,924	0.1381	0.0968	0.0708	0.1198	0.1845
LoanTypest (consumer)	445,924	0.0736	0.0826	0.0188	0.0476	0.0969
LoanTypest (agricultural)	445,924	0.0771	0.1270	0.0000	0.0110	0.1003

Note: Variable definitions are in Appendix A. All bank-specific continuous variables have been winsorized at the 1st and 99th percentiles (except SIZE_t-1 which is expressed as natural logarithm, while BRANCHDIV_t is expressed in percentage).

Table 3. Relation between disaster risk and loan loss provisions.

	Column (1)	Column (2)	Column (3)
	LLPt	LLPt	LLPt
Constant	0.0019	0.0016	−0.4701***
	[0.07]	[0.06]	[−9.42]
DRt	0.0027***	0.0027***	0.0021***
	[9.78]	[9.81]	[6.56]
ΔNPAt+1	1.7317***	1.7224***	1.2800***
	[19.13]	[19.10]	[14.48]
ΔNPAt	3.5484***	3.4997***	3.1288***
	[34.19]	[4.69]	[30.73]
ΔNPAt-1	4.5013***	4.4932***	4.0273***
	[48.17]	[48.24]	[44.11]
ΔNPAt-2	3.7364***	3.7295***	3.3793***
	[43.70]	[43.68]	[39.90]
ΔLOANSt	−0.0113	−0.0074	−0.1158***
	[−0.88]	[−0.57]	[−9.61]
EBTLLPt	5.3512***	5.3855***	6.9741***
	[19.02]	[19.14]	[21.42]
COt-1	49.9029***	49.7603***	40.5223***
	[69.77]	[69.54]	[64.39]
TIER1t-1	0.0810***	0.0834***	0.3451***
	[3.59]	[3.69]	[9.15]
ALLOWANCEt-1	1.4151***	1.3730***	0.5259***
	[10.03]	[9.71]	[3.07]
SIZEt-1	0.0005	0.0003	0.0216***
	[0.45]	[0.33]	[6.53]
BRANCHDIVt	0.0002***	0.0002***	0.0002***
	[7.61]	[7.63]	[3.60]
Loan Types Controls	Yes	Yes	Yes
Time Fixed Effects	Year-Quarter	Year-Quarter × ΔNPAt	Year-Quarter
State-County Fixed Effects	Yes	Yes	No
Bank Fixed Effects	No	No	Yes
Observations	445,924	445,924	445,924
R^2	0.317	0.319	0.358

Note: This table reports estimation results for the following model relating disaster risk (DR_t) to loan loss provisions (LLP_t): $\mathrm{LL}{P}_t={\beta }_0+{\beta }_{1}D{R}_t+\mathrm{\Sigma }{\beta }_j\mathrm{Controls}+\mathrm{\Sigma }{\beta }_i\mathrm{FixedEffects}+{\epsilon }_t$

Variable definitions are in Appendix A. All bank-specific continuous variables are winsorized at the 1st and 99th percentiles (except SIZE_t-1 which is expressed as natural logarithm, while BRANCHDIV_t is expressed in percentage). *, **, and *** indicate significance at the 0.10, 0.05, and 0.01 levels (two-tailed), respectively. t-statistics are presented in parentheses and are based on standard errors clustered at the bank level.

Figure 2 Counties affected by hurricanes and hurricane Katrina during the third quarter of 2005.

Note: Figure 2 presents a map chart of counties affected by hurricane Katrina (colored in light gray, 179 counties, 21,655 bank-quarter observations), counties not affected by hurricane Katrina without hurricane events in the FEMA dataset during the 10 years prior to Hurricane Katrina (colored in black, 2367 counties, 355,414 bank-quarter observations), and counties not affected by hurricane Katrina with at least 1 hurricane event in the FEMA dataset during the 10 years prior to Hurricane Katrina (colored in dark gray, 689 counties, 68,855 bank-quarter observations).

Table 4. Relation between disaster risk and loan loss provisions: difference-in-differences approach.

	Column (1)		Column (2)
	LLPt		LLPt
Constant	−0.4534***	[−8.77]	−0.4505***	[−8.83]
DRt	0.0015***	[3.11]	0.0013***	[2.82]
DRt x POSTt	−0.0000	[−0.01]	0.0000	[0.03]
HURRICANE1t	0.0080	[0.41]
DRt x HURRICANE1t	−0.0000	[−0.03]
POSTt x HURRICANE1t	−0.0055	[−0.77]
DRt x POSTt x HURRICANE1t	0.0033***	[2.59]
HURRICANE2t			−0.0088	[−0.15]
DRt x HURRICANE2t			0.0087***	[2.97]
POSTt x HURRICANE2t			−0.1340***	[−4.57]
DRt x POSTt x HURRICANE2t			0.0180***	[5.38]
ΔNPAt+1	1.2825***	[14.08]	1.2757***	[14.01]
ΔNPAt	3.1272***	[29.89]	3.1188***	[29.84]
ΔNPAt-1	4.0313***	[43.13]	4.0227***	[43.04]
ΔNPAt-2	3.3954***	[39.28]	3.3876***	[39.19]
ΔLOANSt	−0.1105***	[−8.97]	−0.1102***	[−8.95]
EBTLLPt	6.9914***	[20.72]	6.9711***	[20.68]
COt-1	40.2954***	[62.48]	40.1822***	[62.31]
TIER1t-1	0.3347***	[8.60]	0.3325***	[8.56]
ALLOWANCEt-1	0.5476***	[3.05]	0.5359***	[2.99]
SIZEt-1	0.0215***	[6.33]	0.0215***	[6.34]
BRANCHDIVt	0.0002***	[3.31]	0.0002***	[3.35]
Loan Types Controls	Yes		Yes
Time Fixed Effects	Year-Quarter		Year-Quarter
Bank Fixed Effects	Yes		Yes
Observations	424,269		424,269
R^2	0.360		0.360

Note: This table reports estimation results for the following difference-in-differences model: $\begin{array}{rl}\mathrm{LL}{P}_t& ={\beta }_0+{\beta }_{1}D{R}_t+{\beta }_2\mathrm{POS}{T}_t+{\beta }_3\mathrm{HURRICAN}{E}_t+{\beta }_{4}D{R}_t\mathrm{xPOS}{T}_t\\ & +{\beta }_{5}D{R}_t\mathrm{xHURRICAN}{E}_t+{\beta }_6\mathrm{POS}{T}_t\mathrm{xHURRICAN}{E}_t\\ & +{\beta }_{7}D{R}_t\mathrm{xPOS}{T}_t\mathrm{xHURRICAN}{E}_t+\mathrm{\Sigma }{\beta }_j\mathrm{Controls}+\mathrm{\Sigma }{\beta }_i\mathrm{Fixed}\mathrm{Effects}+{\epsilon }_t\end{array}$

Variable definitions are in Appendix A. All bank-specific continuous variables are winsorized at the 1st and 99th percentiles (except SIZE_t-1 which is expressed as natural logarithm, while BRANCHDIV_t is expressed in percentage). The main effect of POST is absorbed by the fixed effects. *,**, and *** indicate significance at the 0.10, 0.05, and 0.01 levels (two-tailed), respectively. t-statistics are presented in parentheses and are based on standard errors clustered at the bank level.

Table 5. Relation between additional measures of disaster risk and loan loss provisions.

	Column (1)	Column (2)	Column (3)	Column (4)
	LLPt	LLPt	LLPt	LLPt
Constant	−0.4746***	−0.4661***	−0.4611***	−0.4701***
	[−9.52]	[−9.33]	[−9.21]	[−9.42]
DR_MULTICOUNTYt	0.0028***
	[7.95]
DR_FREQUENTt		0.0017***		0.0020***
		[5.30]		[5.98]
DR_RAREt			0.0017***	0.0023***
			[3.01]	[4.08]
Bank & Loan Types Controls	Yes	Yes	Yes	Yes
Time Fixed Effects	Year-Quarter	Year-Quarter	Year-Quarter	Year-Quarter
Bank Fixed Effects	Yes	Yes	Yes	Yes
Observations	445,924	445,924	445,924	445,924
R^2	0.358	0.358	0.358	0.358

Note: This table reports estimation results for the following model relating additional measures of disaster risk, namely multicounty disaster risk (DR_MULTICOUNTY_t), frequent event disaster risk (DR_FREQUENT_t), and rare event disaster risk (DR_RARE_t), to loan loss provisions (LLP_t): $\begin{array}{rl}\mathrm{LL}{P}_t& ={\beta }_0+{\beta }_1(\mathrm{DR}\mathrm{\_MULTICOUNT}{Y}_t|\mathrm{DR}\mathrm{\_FREQUEN}{T}_t|\mathrm{DR}\mathrm{\_RAR}{E}_t)\\ & +\mathrm{\Sigma }{\beta }_j\mathrm{Controls}+\mathrm{\Sigma }{\beta }_i\mathrm{FixedEffects}+{\epsilon }_t\end{array}$

Variable definitions are in Appendix A. DR is measured using the disaster risk at branch level (column 1), frequent events (column 2), and rare events (column 3). All bank-specific continuous variables are winsorized at the 1st and 99th percentiles (except SIZE_t-1 which is expressed as natural logarithm, while BRANCHDIV_t is expressed in percentage). *, **, and *** indicate significance at the 0.10, 0.05, and 0.01 levels (two-tailed), respectively. t-statistics are presented in parentheses and are based on standard errors clustered at the bank level.

Table 6. Relation between disaster risk and loan loss provisions: cross-sectional tests based on banks’ characterisctis.

	Column (1)	Column (2)	Column (3)	Column (4)	Column (5)	Column (6)
	LLPt	LLPt	LLPt	LLPt	LLPt	LLPt
Constant	−0.8092***	−0.3364***	−0.5595***	−0.3215***	−0.2936**	−3.3285***
	[−7.92]	[−2.64]	[−6.19]	[−2.87]	[−2.31]	[−13.25]
DRt	−0.0001	0.0027***	0.0056***	0.0015**	0.0005	0.0053***
	[−0.10]	[4.19]	[7.39]	[2.16]	[0.64]	[6.48]
Cross-sectional sample	Small	Large	Simple	Complex	Unconcentrated	Concentrated
Difference in DRt	0.0028***		−0.0041**		0.0048***
Bank & Loan Types Controls	Yes	Yes	Yes	Yes	Yes	Yes
Time Fixed Effects	Year-Quarter	Year-Quarter	Year-Quarter	Year-Quarter	Year-Quarter	Year-Quarter
Bank Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes
Observations	111,484	111,482	111,481	111,489	111,481	111,481
R^2	0.215	0.519	0.374	0.432	0.396	0.418

Note: This table reports estimation results for the following model relating disaster risk (DR_t) to loan loss provisions (LLP_t) for three subsamples based on asset size, bank complexity, and loan portfolio concentration: $\mathrm{LL}{P}_t={\beta }_0+{\beta }_{1}D{R}_t+\mathrm{\Sigma }{\beta }_j\mathrm{Controls}+\mathrm{\Sigma }{\beta }_i\mathrm{FixedEffects}+{\epsilon }_t$

We define small (large) banks if the total asset is lower (higher) than the 25th (75th) percentile of the distribution. We define simple (complex) banks if the non-interest income over total income is lower (higher) than the 25th (75th) percentile of the distribution. We define unconcentrated (concentrated) banks if the standard deviation of loan categories is lower (higher) than the 25th (75th) percentile of the distribution. Variable definitions are in Appendix A. All bank-specific continuous variables are winsorized at the 1st and 99th percentiles (except SIZE_t-1 which is expressed as natural logarithm, while BRANCHDIV_t is expressed in percentage). *, **, and *** indicate significance at the 0.10, 0.05, and 0.01 levels (two-tailed), respectively. t-statistics are presented in parentheses and are based on standard errors clustered at the bank level.

Table 7. Relation between disaster risk, current loan loss provisions and risk management.

	Column (1)	Column (2)	Column (3)	Column (4)
	COt+1, t+4	COt+1, t+4	RISK t+1, t+4	RISK t+1, t+4
Constant	−0.0059***	−0.0056***	−2.9145***	−2.8977***
	[−14.51]	[−7.24]	[−14.28]	[−7.70]
LLPt	0.0018***	0.0015***	0.4571***	0.4990***
	[34.90]	[8.17]	[22.90]	[9.69]
DRt	0.0000***	0.0000	−0.0029	0.0009
	[2.94]	[0.37]	[−1.33]	[0.25]
LLPt x DRt	0.0000	0.0001**	0.0116***	0.0007
	[0.52]	[1.98]	[3.31]	[0.09]
Events in t+1	No	Yes	No	Yes
Bank & Loan Types Controls	Yes	Yes	Yes	Yes
Time Fixed Effects	Year-Quarter	Year-Quarter	Year-Quarter	Year-Quarter
Bank Fixed Effects	Yes	Yes	Yes	Yes
Observations	396,838	35,505	396,465	35,462
R^2	0.554	0.632	0.550	0.643

Note: This table reports estimation results for the following models of the moderating effect of disaster risk (DR_t) on the relation between loan loss provisions (LLP_t), future charge-offs (CO_{t+1, t+4}) and future risk-taking (RISK_t+1): $\begin{array}{rl}C{O}_{t+1,t+4}& ={\beta }_0+{\beta }_1\mathrm{LL}{P}_t+{\beta }_{2}D{R}_t+{\beta }_3\mathrm{LL}{P}_t\times D{R}_t+\mathrm{\Sigma }{\beta }_j\mathrm{Control}{s}_j+\mathrm{\Sigma }{\beta }_i\mathrm{FixedEffect}{s}_i+{\epsilon }_t\end{array}$ $\begin{array}{rl}\mathrm{RIS}{K}_{t+1}& ={\beta }_0+{\beta }_1\mathrm{LL}{P}_t+{\beta }_{2}D{R}_t+{\beta }_3\mathrm{LL}{P}_t\times D{R}_t+\mathrm{\Sigma }{\beta }_j\mathrm{Control}{s}_j+\mathrm{\Sigma }{\beta }_i\mathrm{FixedEffect}{s}_i+{\epsilon }_t\end{array}$

Variable definitions are in Appendix A. All bank-specific continuous variables are winsorized at the 1st and 99th percentiles (except SIZE_t-1 which is expressed as natural logarithm, while BRANCHDIV_t is expressed in percentage). *, **, and *** indicate significance at the 0.10, 0.05, and 0.01 levels (two-tailed), respectively. t-statistics are presented in parentheses and are based on standard errors clustered at the bank level.

Table

Name	Description	Source
Loan Loss Provisions (LLPt)	Provision for credit losses in quarter t divided by net loans and leases in quarter t-1.	FDIC: ELNATQ, LNLSNET.
Future net charge-offs on loan (COt+1, t+4)	Average of net charge-offs from quarter t+1 to quarter t+4, divided by net loans and leases in quarter t.	FDIC: NTLNLSQ, LNLSNET.
Future risk-taking (RISKt+1, t+4)	Average z-score from quarter t+1 to quarter t+4. We measure z-score as the natural logarithm of [(ROA + CAP_REG) / SD_ROA], where ROA is income before extraordinary items and provision for credit losses divided by total assets, CAP_REG is total regulatory capital ratio (Tier-1), SD_ROA is the standard deviation of ROA on four quarters rolling window. We multiply the score by −1 so that higher z-score implies higher risk-taking.	FDIC: IBEFXTRQ, ELNATQ, RBCT1J, ASSET.
Disaster Risk (DRt)	Number of natural disasters declared as major disasters over a 15-year span by the Federal Emergency Management Agency in each county-quarter.	Federal Emergency Management Agency
Multicounty Disaster Risk (DR_MULTICOUNTYt)	Weighted average measure of disaster risk using the number of branches in each county as weights.	Federal Emergency Management Agency; Bank Regulatory: Federal Deposit Insurance Corporation, Summary of Deposits.
Frequent Event Disaster Risk (DR_FREQUENTt)	The number of natural disasters related to frequent events declared as major disasters over a 15-year span by the Federal Emergency Management Agency in each county-quarter. We define frequent natural events as events that occurred with a frequency equal or higher than the median of the events frequency in each county.	Federal Emergency Management Agency
Rare Event Disaster Risk (DR_RAREt)	The number of natural disasters related to rare events declared as major disasters over a 15-year span by the Federal Emergency Management Agency in each county-quarter. We define rare natural events as events that occurred with a frequency lower than the median of the events frequency in each county.	Federal Emergency Management Agency
Post Hurricane Katrina (POSTt)	Indicator variable that takes the value of 1 after Hurricane Katrina strike (i.e. third quarter of 2005), 0 otherwise.	Federal Emergency Management Agency
Hurricane hazard area (HURRICANE1t)	Indicator variable that takes the value of 1 if a bank is headquartered in a county that registered at least one hurricane event in the FEMA dataset in the ten years before Hurricane Katrina (i.e. third quarter of 2005), 0 otherwise.	Federal Emergency Management Agency
Hurricane hazard area (HURRICANE2t)	Indicator variable that takes the value of 1 if a bank is headquartered in a county that registered frequent hurricane strikes in the FEMA dataset in the ten years before Hurricane Katrina (i.e. third quarter of 2005), 0 otherwise. We define frequent hurricane strikes counties as counties in the top quartile of the distribution of the number of hurricane event.	Federal Emergency Management Agency
Change in nonperforming assets (ΔNPAt-2; ΔNPAt-1; ΔNPAt; ΔNPAt+1)	Change in assets in nonaccrual status from quarter t-1 to quarter t, divided by net loans and leases in quarter t-1.	FDIC: NAASSET, LNLSNET.
Change in total loans and leases (ΔLOANSt)	Change in net loans and leases from quarter t-1 to quarter t, divided by net loans and leases in quarter t-1.	FDIC: LNLSNET.
Earnings before taxes and loan and leases losses (EBTLLPt)	Sum of income before extraordinary items and provision for credit losses, both in quarter t, divided by net loans and leases in quarter t-1.	FDIC: IBEFXTRQ, ELNATQ, LNLSNET.
Net charge-offs on loan (COt-1)	Past four quarters rolling average of net charge-offs in quarter t divided by net loans and leases in previous quarter.	FDIC: NTLNLSQ, LNLSNET.
Tier1 capital ratio (TIER1t-1)	Tier one (core) capital in quarter t-1 divided by total assets in quarter t-2.	FDIC: RBCT1J, ASSET.
Allowance for loan and leases losses (ALLOWANCEt-1)	Loan and leases loss allowance in quarter t-1 divided by net loans and leases in quarter t-2.	FDIC: LNATRES, LNLSNET.
Size (SIZEt-1)	Natural logarithm of total assets in quarter t-1.	FDIC: ASSET.
Branch diversification (BRANCHDIVt)	Number of bank branches outside the county where the bank is headquartered divided by the total number of bank branches in year t.	FDIC: STCNTYBR, STCNTY.
Loan portfolio composition (LoanTypest)	Residential, commercial, consumer and agricultural loans in quarter t divided by net loans and leases in quarter t.	FDIC: LNRE, LNCI, LNCON, LNAG, LNLSNET.

Data Availability Statement

The data are available from the sources identified in the text.