README FILE FOR REPLICATION OF RESULTS IN “BANKS' STRATEGIC RESPONSES TO SUPERVISORY COVERAGE: EVIDENCE FROM A NATURAL EXPERIMENT,” BY IVAN IVANOV, BEN RANISH AND JAMES WANG, FORTHCOMING IN THE JOURNAL OF MONEY, CREDIT AND BANKING.

OUR RESULTS CAN BE REPLICATED FROM THE DATA SETS AND CODE DESCRIBED BELOW.

IVAN IVANOV, BEN RANISH AND JAMES WANG, FEBRUARY 2022.

**********************************
******** DATA SOURCES USED *******
**********************************

*Primary Sources

*Shared National Credit (SNC): This confidential supervisory database contains information about syndicated credits included in the scope of each exam. This information includes the complete set of lenders and their share of the loan, the agent lenders' risk assment of the loan, loan characteristics, basic borrower characteristics, and summary of the examiners' assessment of the loan. We use much of this information in our analyses. Shared National Credit data can only be accessed by U.S. banking regulators, used only for approved purposes, and personally identifiable information from the data may not be publicly disclosed.

*Dealscan: This commercially available database (which may be accessible through WRDS) contains information on syndicated credits from public (e.g. SEC) filings. Information about the lenders may be incomplete, though the data do include many loan characteristics including pricing (which is not present in Shared National Credit). We use DealScan data primarily in Section 4.2 of our paper.

*Supplemental Sources

*Moody's: We use long-term issuer credit ratings from this commercially available source as one of our measures of credit risk.

*Compustat and CapitalIQ: We use these commercially available data to determine which borrowers are public companies.

*Bank Holding Company Reports (FR Y-9C, Y-7): We use data from these regulatory reports to measure bank lenders' leverage and size. The particular data items we use are public, but Y-7 data (covering the U.S. presence of foreign banking organizations) are not published and available only upon request. FR Y-9C data are accessible through the National Information Center website (https://www.ffiec.gov/NPW). 

* Tobias Berg's website: We use a total cost of borrowing measure developed by and available on the author's website (https://sites.google.com/view/tobias-berg/startseite/data-and-code).

*SNC-DealScan link: A mapping between credits in DealScan and deals in SNC as used in Bord and Santos (2015): http://www.jstor.org/stable/24500930

*We can share only the Y-9C dataset we use upon request, but the other datasets may only be shared with explicit permission from the data providers, and the Shared National Credit data may only be accessed by U.S. banking regulators (who participate in the Shared National Credit exams).

************************************
******** REQUIRED DATA FILES *******
************************************

*The code provided use the data files described below.

*snc_credits.dta - Each observation in this confidential SNC data set provides characteristics of a single SNC credit at a single SNC exam. The characteristics given for each credit include the identity of the (lead) bank where the credit was reviewed, the identity of the borrower, borrower characteristics (location, industry), credit characteristics (loan type, purpose, origination and maturity dates, committed and utilized balance), and the SNC examiner and lead banks' ratings for the credit.

*snc_lenders.dta - Each observation in this confidential SNC data set provides detail on a given syndicated loan participation, i.e. the lender's identity, location, type (e.g. domestic bank, nonbank, etc), regulator (if applicable) and share of the credit's commitment and utilization. This data set maps many-to-one to snc_credits.

*dealscan_for_snc_merge.dta - This table is constructed from a merge of several DealScan tables. Each observation represents a DealScan loan facility (one or more of which comprise a syndicated credit) when publicly observed (issuance or major renegotiation). Data include the borrower ID as well as characteristics of the facility including line/loan structure, commitment amounts, pricing, loan purpose and whether there are collateral or covenants.

*deal_id.dta - This table maps SNC credits many-to-one into SNC deals, as "deals" are the level at which the SNC exams are conducted and credits are included/excluded. A deal consists of one or more credits that at any point in time have the same (i) borrower, (ii) agent/lead bank and (iii) origination date.

*snc_clean.dta - This is our primary dataset used for SNC-based analysis. It consists of lender information (from snc_lenders.dta) and credit information (from snc_credits.dta) aggregated to the deal level (using deal_id.dta). In addition, this table includes a few credit quality proxies: the borrower's imputed external credit rating (using Moody's credit rating data in conjunction with the lead banks' internal credit rating scale), and an estimated probability of downgrade and increase in classified assets associated with examination. The table includes a public obligor dummy (Compustat/CapitalIQ) and lender size and leverage data for bank lenders (from Y-9C and Y-7 regulatory reports).

*dealscan_data_for_reg.dta - This is our primary dataset used for the DealScan analysis, consisting of data from (i) dealscan_for_snc_merge (ii) snc_clean and (iii) Tobias Berg's total cost of borrowing (TCB) measure. The DealScan data are merged with SNC through use of the "SNC-DealScan" link dataset used in Bord and Santos (2015).

*data_for_bank_level_regs.dta - This table is derived from snc_clean.dta, and reports for each bank lender and year the number and amount of deals held and newly originated, by number of syndicate participants, along with lender size and leverage.

The authors are able to share the code used to assemble these datasets from the underlying data sources upon request. However, certain lines may require redaction to preserve micro-data confidentiality. Also note: since work on this project began, the SNC database schema has changed. While snc_credits and snc_lenders were once the output of simple "SELECT" queries on the SQL database, table names and contents may have been rearranged.

********************************
******** REQUIRED CODE   *******
********************************

*All code is executed using STATA 15.1 MP

*Run_analysis.do - sets file paths used throughout the code and downloads the STATA packages need to run the code.

*diff_in_diff_plots.do - generates the difference-in-differences plots (Figure 1)

*summary_stat_tables.do - generates the summary statatistics tables (Tables 1 and 2)

*lender_transition.do - runs the analysis of transition probabilities between number of lenders (Table 3)

*deal_level_regressions.do - runs deal-level difference-in-differences regressions of deal exit and credit quality on the number of deal lenders (Tables 4 and 5)

*analysis_of_twolender_borrowers.do - runs our borrower-level regressions of deal terms on borrowers' use of two/three lender deals (Table 6)

*bank_level_reg.do - runs our lender-level regressions of lenders' holdings of two/three lender deals (Table 7)

*Ivan Ivanov, Ben Ranish and James Wang, January 2022