References
Adams-Kane, Jonathon. 2018. “Commercial Real Estate: How Vulnerable are U.S. Banks?” Milken Institute Santa Monica, CA.
Cole, R., and L. White. 2012. “Déjà Vu All Over Again: The Causes of U.S. Commercial Bank Failures This Time Around.” Journal of Financial Services Research 42 (1), 5–29.
D’Erasmo, Pablo. 2019. “Banking Trends: Estimating Today’s Commercial Real Estate Risk.” Economic Insights 4(1): 9–15.
Dingle, J. I., and B. Neiman. 2020. “How Many Jobs Can be Done at Home?” NBER Working Paper 26948, National Bureau of Economic Research, Cambridge, MA.
Federal Reserve. 2021. Coronavirus Disease 2019 (COVID-19) website. https://www.federalreserve.gov/covid-19.htm. Access Date: May 3, 2021.
FDIC Banker Resource Center. https://www.fdic.gov/resources/bankers/credit/commercial-real-estate-lending/#:~:text=Commercial%20real%20estate%20(CRE)%20lending,owner%20or%20a%20related%20party. Access Date: April 14, 2021.
International Monetary Fund (IMF). 2020. United States – Financial System Stability Assessment. IMF Country Report No. 20/242.
International Monetary Fund (IMF). 2021. “Commercial Real Estate: Financial Stability Risks During the COVID-19 Crisis and Beyond” in Global Financial Stability Report. Washington, DC, April. https://www.imf.org/en/Publications/GFSR/Issues/2021/04/06/global-financial-stability-report-april-2021 Access Date: April 14, 2021.
Hirtle, B., A. Kovner, J. Vickery, and M. Bhanot. 2015. “Assessing Financial Stability: The Capital and Loss Assessment under Stress Scenarios (CLASS) Model.” Staff Report No. 663, Federal Reserve Bank of New York.
Kashyap, A., and J. Stein. 2000. “What Do One Million Observations on Banks Have to Say About the Transmission of Monetary Policy.” American Economic Review 90(3): 407–28.
Khwaja, A. I., and A. Mian. 2008. “Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market.” American Economic Review 98(4): 1413–42.
Annex. Data and Empirical Strategy
Three main data sources have been used to obtain information for the empirical analyses: (i) Federal Financial Institutions Examination Council 031 and 051 Report Forms (Call Reports) for detailed bank-level data at a quarterly frequency on commercial real estate (CRE) loans,14 several outcome indicators and bank controls (to be defined in this Annex), and the location of bank headquarters at the zip code level; (ii) MSCI Real Estate for average quarterly CRE prices in the metropolitan statistical area (MSA) where the bank headquarters are located;15 and (iii) the Federal Deposit Insurance Corporation (FDIC) Deposit Survey to measure the geographical concentration of banks. These databases are matched using unique bank identifiers. The sample period is 2001:Q1–2020:Q3, and includes a total of 10,796 reporting banks (5,116 banks in 2020:Q3).
The analysis of the impact of a change in CRE prices on banks’ performance is based on the following model:
where
k≥1 capture potential lags in the transmission and k=8 quarters is used as the baseline. Controls are bank capital ratio, liquidity ratio, and (log) total asset size, all measured at t-k, the interaction of these variables with
The identification is through the within-MSA variation in bank ex-ante CRE loan exposures (at a given quarter). Similar to Khwaja and Mian (2006), the sample is restricted to MSAs with multiple banks, but the results are strongly robust to including single-bank MSAs in the analyses (there are four MSAs with a single bank headquarters). Bank fixed effects and further controls help to better quantify the effect of CRE loan exposures.18
There are a few data limitations. First, CRE price indices are available only for 69 major MSAs. 19 In the baseline estimations, banks located in these MSAs are used, which, on average, cover more than 50 percent of total banking sector assets. The results are robust, using an expanded sample that includes all banks. 20 Second, bank CRE loans by segments such as loans secured by office, retail, industrial, or lodging properties, are not available in Call Reports. Therefore, average CRE prices—as reported by the MSCI for each MSA—are used. A further limitation is that locations of underlying properties is not known. In the analyses, banks (or loan performance) are assumed to be exposed to CRE price changes in the MSAs where the bank headquarters is located. This limitation, however, may not be as restrictive as it may appear since most banks in the US are small and tend to have geographically concentrated loan portfolios.
Table A1 reports the definition and summary statistics of the key variables used in the analyses.
Definition and Summary Statistics
Definition and Summary Statistics
| Dependent Variables: | Definition | Mean | Std | Obs. | Independent Variables: | Definition | Mean | Std | Obs. |
|---|---|---|---|---|---|---|---|---|---|
| CRE NPL Ratio (in ppt) | Nonperforming CRE loans (90+ days overdue)-to-total CRE loans ratio | 1.56 | 3.45 | 94521 | CRE Loans to Total Assets | The sum of construction, land development, and other land loans; loans secured by multi-family (5 or more) residential properties; and loans secured by nonfarm nonresidential properties-to-total assets ratio | 0.30 | 0.19 | 94521 |
| CRE Charge- Off Rate (net of recovery rate) | Charged off CRE loans (net of recoveries)-to total CRE loans ratio | 0.00 | 0.01 | 90447 | Bank Total Assets (in logs) | Book value of total assets of a bank | 12.73 | 1.56 | 94521 |
| Change in Pre-Provision Net Revenues | 8-quarter log change in annual pre-provision net revenues | 0.15 | 0.75 | 94521 | Bank Liquid Assets-to-Total Assets Ratio | Cash and balances due from depository institutions + securities + federal funds sold and securities purchased under agreements to resell (to total assets) | 0.26 | 0.17 | 94521 |
| Change in Regulatory Capital | 8-quarter log change in regulatory capital | 0.17 | 0.24 | 93462 | Bank Equity Capital-to-Total Assets | Ratio of total bank equity capital to total assets | 0.13 | 0.13 | 94521 |
| Change in CRE Prices | 8-quarter log change in MSA-level average CRE prices | 0.09 | 0.16 | 121187 |
Definition and Summary Statistics
| Dependent Variables: | Definition | Mean | Std | Obs. | Independent Variables: | Definition | Mean | Std | Obs. |
|---|---|---|---|---|---|---|---|---|---|
| CRE NPL Ratio (in ppt) | Nonperforming CRE loans (90+ days overdue)-to-total CRE loans ratio | 1.56 | 3.45 | 94521 | CRE Loans to Total Assets | The sum of construction, land development, and other land loans; loans secured by multi-family (5 or more) residential properties; and loans secured by nonfarm nonresidential properties-to-total assets ratio | 0.30 | 0.19 | 94521 |
| CRE Charge- Off Rate (net of recovery rate) | Charged off CRE loans (net of recoveries)-to total CRE loans ratio | 0.00 | 0.01 | 90447 | Bank Total Assets (in logs) | Book value of total assets of a bank | 12.73 | 1.56 | 94521 |
| Change in Pre-Provision Net Revenues | 8-quarter log change in annual pre-provision net revenues | 0.15 | 0.75 | 94521 | Bank Liquid Assets-to-Total Assets Ratio | Cash and balances due from depository institutions + securities + federal funds sold and securities purchased under agreements to resell (to total assets) | 0.26 | 0.17 | 94521 |
| Change in Regulatory Capital | 8-quarter log change in regulatory capital | 0.17 | 0.24 | 93462 | Bank Equity Capital-to-Total Assets | Ratio of total bank equity capital to total assets | 0.13 | 0.13 | 94521 |
| Change in CRE Prices | 8-quarter log change in MSA-level average CRE prices | 0.09 | 0.16 | 121187 |
Scenario Analysis. Using the estimated values of αk and βk in equation (1), we calculate
A forward-looking provisioning rule in line with supervisory practices is assumed, where provision expenses are calculated based on Allowance for Loan and Lease Losses (ALLL) due to CRE loan exposure equal to four quarters of projected net CRE loan charge-offs. Total CRE loan balances are assumed to decline by 2 percent quarterly, which is in line with the CRE charge-off rates observed during the global financial crisis.21
In the next step, by how much banks’ capital could go under stress is calculated. In particular, projected provision expenses and losses in pre-provision net revenues are divided by total risk-weighted assets, where total risk-weighted assets are assumed to be equal to the last historical observation (2020: Q3).
Incorporating macroeconomic effects into the picture. Macroeconomic effects are incorporated into the picture by including location-specific factors (proxied by MSA x time fixed effects) in the forecast horizon. These fixed effects are assumed to follow a path observed during the global financial crisis.
Figure A1 assesses the information content of MSA x time fixed effects (estimated for the two specifications, that is, the CRE charge-off rate and PPNR, and for different horizons). It shows that these fixed effects are significantly correlated with MSA-level change in CRE prices. The correlations are weaker for the PPNR specification (as banks’ overall PPNR could potentially be more correlated with macroeconomic conditions than the CRE charge-off rate), and stronger for the CRE charge-off rate specification.
Correlation of Time-Varying Location-Specific Factors with Changes in MSA-Level CRE Prices
Citation: Global Financial Stability Notes 2021, 001; 10.5089/9781513578286.065.A999
Sources: Call Reports; and author calculations.Note: The figure shows the correlation of estimated MSA x time fixed effects with MSA-level change in CRE prices (for different horizons). The correlation coefficients are statistically significant at 10 percent or lower (except horizon 6 for the PPNR specification). CRE = commercial real estate; PPNR = pre- provision net revenue.Correlation of Time-Varying Location-Specific Factors with Changes in MSA-Level CRE Prices
Citation: Global Financial Stability Notes 2021, 001; 10.5089/9781513578286.065.A999
Sources: Call Reports; and author calculations.Note: The figure shows the correlation of estimated MSA x time fixed effects with MSA-level change in CRE prices (for different horizons). The correlation coefficients are statistically significant at 10 percent or lower (except horizon 6 for the PPNR specification). CRE = commercial real estate; PPNR = pre- provision net revenue.Correlation of Time-Varying Location-Specific Factors with Changes in MSA-Level CRE Prices
Citation: Global Financial Stability Notes 2021, 001; 10.5089/9781513578286.065.A999
Sources: Call Reports; and author calculations.Note: The figure shows the correlation of estimated MSA x time fixed effects with MSA-level change in CRE prices (for different horizons). The correlation coefficients are statistically significant at 10 percent or lower (except horizon 6 for the PPNR specification). CRE = commercial real estate; PPNR = pre- provision net revenue.I would like to thank Andrea Deghi, Fabio Natalucci, Mahvash S. Qureshi, and Jérôme Vandenbussche for very helpful suggestions and comments, and Zhi Ken Gan for excellent assistance with the data.
Banks are the largest providers of CRE debt financing in the US and globally, but nonbank financial institutions such as pension funds, insurers and investment funds also play an important role. The share of total outstanding CRE debt held by banks is 54 percent in the US (and about 70 percent in Europe and Asia), while that held by nonbank financial institutions is 24 percent in the US (and less than 20 percent in Europe and in Asia; IMF, 2021).
Lower CRE prices–which generally coincide with lower economic activity and reduced cash flow for firms–could affect financial stability by adversely impacting the credit quality of borrowers. Concurrently, loan-to-value (LTV) constraints may also be triggered, especially when the price drop is steep and/or for loans with already high LTV ratios. In extreme cases, a loan balance could even be worth more than the property value, making it excessively risky for lenders to retain the exposure, in particular if the price drop appears persistent. Eventually, following an increase in non-performing CRE loans and lower revenues, bank capital could decline (D’Erasmo, 2019).
Identifying how changes in CRE prices could affect bank capital ideally requires supervisory loan-level data and a real-time appraisal value of real estate assets backing each loan. Such granular data is not publicly available, and the analysis in this note relies on bank-level data on CRE loan exposures (Call Reports) and Metropolitan Statistical Area (MSA)-level data on average CRE prices (obtained from the MSCI).
See IMF (2020) for a detailed assessment of US financial system stability and policy recommendations to strengthen financial system resilience.
The Federal Deposit Insurance Corporation (FDIC) Banker Resource Center defines CRE lending as “… acquisition, development, and construction financing and the financing of income-producing real estate.,”, where “(I)ncome-producing real estate includes real estate held for lease to third parties and nonresidential real estate that is occupied by its owner or a related party.,” For a detailed overview of the US banking sector’s exposure to CRE market developments, see Adams-Kane (2018).
In the context of the global financial crisis, Cole and White (2012) find that real estate construction and development loans, commercial mortgages, and multi-family mortgages were significantly associated with a higher likelihood of bank failure, whereas residential single- family mortgages did not play a significant role.
Based on FDIC Survey of Deposits database, more than 90 percent of small banks (in particular, banks with total assets that never exceed $5 billion throughout our sample period) have offices in a single state and in at most two metropolitan statistical areas (MSAs). In contrast, top 41 banks (that correspond to banks with total assets that exceed $100 billion at least once throughout the sample period) have offices in 9 states or 45 MSAs on average (and 42 states and 253 MSAs at the maximum).
See the Annex for details on the empirical methodology and data.
The third quartile of the distribution of CRE loan exposure corresponds to the total CRE loans-to-total assets ratio of 43 percent. For further details on banks’ CRE loan exposures, see Annex Table A1.
The methodology adopted here broadly follows Hirtle et al. (2015), and is not intended to be strictly comparable to a formal top-down stress test, where the entire bank balance sheet is assumed to be under stress under certain macroeconomic scenarios and given assumptions about tax, dividend and other capital distributions over the forecast horizon. See the Annex for a detailed description of the methodology.
In the left tail (the lowest 5 th percentile) of the distribution of projected capital losses, 90 percent are small banks, 10 percent are medium- sized banks, and 99 percent are community banks.
To incorporate macroeconomic effects into the picture, a path for time-varying locational factors is assumed over the forecast horizon (proxied by fixed effects). See the online annex for further details.
See Fed (2021) for a summary of Federal Reserve’s policy responses to the COVID-19 crisis
Commercial real estate loans throughout the analyses are defined as (i) construction, land development, and other land loans; (ii) loans secured by multi-family (five or more) residential properties; and (iii) loans secured by nonfarm, nonresidential properties (owner-occupied and other nonfarm nonresidential properties).
While MSCI provides CRE price indices across different CRE segments-for example, office, retail, industrial, and lodging-for each MSA at a given quarter, such disaggregated, segment-level data on bank loans is not available. Therefore, average CRE prices reported by the MSCI are used.
See Table A2 for a detailed definition of the variables used in the analyses.
Banks’ exposure to CRE markets is proxied by their outstanding CRE loans (relative to their total assets). In addition to CRE loans, banks’ holdings of CRE-derived securities (CMBS) and real estate may also expose them to fluctuations in CRE prices. To the extent banks’ CRE loan exposures are correlated with their CMBS or real estate holdings, the results should go through.
For instance, as presented in panel 3 of Figure 1, large banks on average have lower CRE loan exposures compared to smaller banks. Controlling for bank size mitigates a potential concern that the identified effect of CRE loan exposure is mismeasured due to its interplay with bank size.
MSAs included in the analyses are the following: Allentown-Bethlehem-Easton, PA-NJ; Anderson, IN; Atlanta-Sandy Springs-Marietta, GA; Austin-Round Rock-San Marcos, TX; Baltimore-Towson, MD; Birmingham-Hoover, AL; Boston-Cambridge-Quincy, MA-NH; Boulder, CO; Bridgeport-Stamford-Norwalk, CT; Cape Coral-Fort Myers, FL; Charlotte-Concord-Gastonia, NC-SC; Chicago-Joliet-Naperville, IL-IN- WI; Cincinnati-Middletown, OH-KY-IN; Columbia, MO; Columbia, SC; Columbia, TN; Columbus, OH; Dallas-Fort Worth-Arlington, TX; Denver-Aurora-Lakewood, CO, NC-SC; Detroit-Warren-Dearborn, MI; Durham-Chapel Hill, NC; Greenville-Anderson-Mauldin, SC; Harrisburg-Carlisle, PA; Hartford-West Hartford-East Hartford, Houston-The Woodlands-Sugar Land, TX; Indianapolis-Carmel, IN; Jacksonville, FL; Kansas City, MO-KS; Knoxville, TN; Lakeland-Winter Haven, FL; Las Vegas-Henderson-Paradise, NV; Los Angeles-Long Beach-Santa Ana, CA; Louisville/Jefferson County, KY-IN; Memphis, TN-MS-AR; Miami-Fort Lauderdale-Pompano Beach, FL; Milwaukee- Waukesha-West Allis, WI; Minneapolis-St. Paul-Bloomington, MN-WI; Naples-Marco Island, FL; Nashville-Davidson--Murfreesboro— Franklin, TN; New Haven-Milford, CT; New Orleans-Metairie-Kenner, LA; New York-Newark-Jersey City, NY-NJ-PA; North Port-Bradenton- Sarasota, FL; Oklahoma City, OK; Orlando-Kissimmee-Sanford, FL; Oxnard-Thousand Oaks-Ventura, CA; Philadelphia-Camden- Wilmington, PA-NJ-DE-MD; Phoenix-Mesa-Glendale, AZ; Pittsburgh, PA; Port St. Lucie, FL; Portland-Vancouver-Hillsboro, OR-WA; Providence-Warwick, RI-MA; Raleigh, NC; Reno, NV; Riverside-San Bernardino-Ontario, CA; Sacramento--Arden-Arcade--Roseville, CA; Salt Lake City, UT; San Diego-Carlsbad-San Marcos, CA; San Francisco-Oakland-Fremont, CA; San Jose-Sunnyvale-Santa Clara, CA; Santa Rosa, CA; Seattle-Tacoma-Bellevue, WA; St. Louis, MO-IL; Stockton-Lodi, CA; Tampa-St. Petersburg-Clearwater, FL; Tucson, AZ; Virginia Beach-Norfolk-Newport News, VA-NC; Washington-Arlington-Alexandria, DC-VA-MD-WV; and Worcester, MA.
The results are qualitatively robust to expanding the sample by using the average of the MSA-level CRE prices within a state- or Census Bureau region, and accordingly, also including banks that are not located in one of these MSAs. The drawback of this approach is the incompatibility within the estimation equation regarding what is meant by location. In particular, “location”-specific CRE prices and location fixed effects would not span the same geographical area.
Assuming a decline in CRE loan balances in the forecast path implies that banks are assumed to be less willing to originate new CRE loans or reluctant to re-finance existing ones, so that in equilibrium, the existing stock of CRE loan balances declines. Assuming constant CRE loan balances over the forecast horizon implies mildly stronger results.
