Credit_COVID19_Canada

This is the code base to accompany the manuscript Consumer Credit Usage in Canada during the Coronavirus Pandemic by Ho, Morin, Paarsch and Huynh in the Canadian Journal of Economics, 2021

Any updates will be available on the GitHub code repository available at the following link: Credit_COVID19_Canada

Data Availability

TransUnion® Data

The primary data source is anonymized consumer credit data from TransUnion®. Data are provided to the Bank of Canada on a monthly basis. Under the contractual agreement with TransUnion, the data are not publicly available. The Bank of Canada does, however, have a process for external researchers to work with these data. The Bank of Canada's Financial System Research Center is a hub for research on household finance. Interested parties, who are Canadian citizens or permanent residents, can contact Jason Allen (Jallen@bankofcanada.ca) or the Managing Director of research Jim MacGee (JMacGee@bankofcanada.ca). Interested parties are asked to submit a project proposal; the proposal is evaluated by senior staff at the Bank of Canada for feasibility; external researchers do not typically have direct access to the data and must work with a Bank of Canada staff member. An exception is if an external collaborator applies for and is granted temporary employee status -- in this case, the external researcher can access the data so long as they have a Bank of Canada affiliation. All research is vetted by Bank of Canada senior staff prior to publication.

Regulatory Filings by the Bank of Canada (formerly E2)

The Bank of Canada’s historical credit aggregates are available in the Banking and Financial Statistics table. Since October 2020, Statistics Canada produces monthly credit aggregates that aligns with the Statistics Canada’s National Balance Sheet Accounts program. These new credit aggregates are available from tables at Statistics Canada: Consumer credit, outstanding balances of selected holders (CANSIM table 10-10-0117-01). For details, see the Bank of Canada's announcement Bank of Canada and Statistics Canada to move to a single set of credit statistics.

The Nilson Report

The set of numbers in some tables are taken directly from the The Nilson Report, Issue 1173, April 2020, and are available here.

2016 Canadian Census

The number of credit-card account holders aged 20 and above and other figures were obtained from Statistics Canada in the table called *Estimates of population (2016 Census and administrative data), by age group and sex for July 1st, Canada, provinces, territories, health regions (2018 boundaries) and peer groups, * Table: 17-10-0134-01.

Instructions:

The workflow proceeds in two stages: one set of instructions outlines the operations to transform the raw data in the TransUnion® database into the datasets that are the inputs for the statistical analysis in the next stage.

Data Manipulation

These procedures were performed on the EDITH 2.0 computing cluster at the Bank of Canada to generate the primary datasets. These scripts are stored in the Code/Data_Prep folder.

1. Run the SLURM script df_ind_bc.slurm, which runs a sequence of Python scripts cr_use_bc_Y1Y2.py, for data covering each two-year period 20Y1-20Y2. It then runs cr_use_bc_combine.py, which generates a temporary parquet file df_ind.parquet. This dataset comprises individual-level data that is sufficient to run the data manipulation for credit-card accounts on the nation-wide sample.

2. Run the script tu_sample_bc.slurm, which runs the script tu_sample_bc.py and generates the dataset tu_sample_bc.csv. This dataset is sufficient to run the analysis of credit-card accounts on the nation-wide sample.

3. Run the script tu_agg_series_bc.slurm, which runs the script tu_agg_series.py and generates the dataset tu_agg_bc.csv. This dataset provides the input for panel (a) of Figure 1: Consumers' Outstanding Balances, 2017-2020 for credit-card accounts on the nation-wide sample.

4. Run the script tu_sample_AB_bc.slurm, which runs the script tu_sample_AB_bc.py and generates the dataset tu_sample_AB_bc.csv. This dataset is sufficient to run the analysis of credit-card accounts on the Alberta sample.

5. Run the script tu_agg_series_AB_bc.slurm, which runs the script tu_agg_series_AB.py and generates the dataset tu_agg_AB_bc.csv. This dataset provides the input for panel (a) of Figure 8: Consumers' Outstanding Balances, Alberta, 2012-2016 for credit-card accounts on the Alberta sample.

6. Run the script TU_vs_Nilson_comp.slurm, which runs the script TU_vs_Nilson_comp.py and generates the dataset TU_vs_Nilson_num_accts.csv. This dataset provides the input for Table A1: Comparison of Accounts at the Credit Agency with Nation-Wide Totals in The Nilson Report.

7. Run the script TU_vs_BoC_comp.slurm, which runs the script TU_vs_BoC_comp.py and generates the dataset TU_vs_BoC_totals.csv. This dataset provides the input for Figure A1.1: Time Series of Aggregate Credit-Card Balances.

8. Run the script TU_vs_StatsCan_comp, which runs the script TU_vs_StatsCan_comp.py and generates the dataset CC_TU_vs_StatsCan.csv. This dataset provides the input for Figure A1.2: Credit Data Coverage for Adults in Canada, by Province.

9. Run the script df_ind_heloc.slurm, which runs a sequence of Python scripts cr_use_heloc_Y1Y2.py, for data covering each two-year period 20Y1-20Y2. It then runs cr_use_heloc_combine.py, which generates a temporary parquet file df_ind.parquet. This dataset comprises individual-level data that is sufficient to run the data manipulation for HELOC accounts on the nation-wide sample.

10. Run the script tu_sample_heloc.slurm, which runs the script tu_sample_heloc.py and generates the dataset tu_sample_heloc.csv. This dataset is sufficient to run the analysis of HELOC accounts on the nation-wide sample.

11. Run the script tu_agg_series_heloc.slurm, which runs the script tu_agg_series.py and generates the dataset tu_agg_heloc.csv. This dataset provides the input for panel (b) of Figure 1: Consumers' Outstanding Balances, 2017-2020 for HELOC accounts on the nation-wide sample.

12. Run the script tu_sample_AB_heloc.slurm, which runs the script tu_sample_AB_heloc.py and generates the dataset tu_sample_AB_heloc.csv. This dataset is sufficient to run the analysis of HELOC accounts on the Alberta sample.

13. Run the script tu_agg_series_AB_heloc.slurm, which runs the script tu_agg_series_AB.py and generates the dataset tu_agg_AB_heloc.csv. This dataset provides the input for panel (b) of Figure 8: Consumers' Outstanding Balances, Alberta, 2012-2016 for HELOC accounts on the Alberta sample.

Datasets

The above operations will produce the following datasets in csv format.

Main datasets

The Data folder must contain four main datasets: tu_sample_bc.csv and tu_sample_heloc.csv for the nation-wide sample, as well as tu_sample_AB_bc.csv and tu_sample_AB_heloc.csv for the sample restricted to the province of Alberta.

tu_sample_bc.csv

This dataset contains observations of credit card balances for consumers in Canada from 2017-2021. It contains the following variables:

1. tu_consumer_id is a 9-digit integer that indicates an individual consumer.
2. Run_Date is a date variable of the form 'YYYY-MM-01' indicating the month in which the data were reported by the bureau. It is the date that represents the last information added to the files, so it contains the statement activity recorded in the previous month.
3. prov is a string that indicates the province of residence of the consumer.
4. homeowner is an indicator that the consumer has ever had a mortgage or a HELOC loan.
5. N_bc is the number of credit card accounts held by a consumer.
6. bc_bal is the consumer's credit-card balance in dollars.

tu_sample_he.csv

This dataset contains observations of HELOC balances for consumers in Canada from 2017-2021. It contains the following variables:

1. tu_consumer_id is 9-digit integer that indicates an individual consumer.
2. Run_Date is a date variable of the form 'YYYY-MM-01' indicating the month in which the data were reported by the bureau.
3. prov is a string that indicates the province of residence of the consumer.
4. homeowner is an indicator that the consumer has ever had a mortgage or a HELOC loan.
5. N_he is the number of HELOC accounts held by a consumer.
6. he_bal is the consumer's HELOC balance in dollars.

tu_sample_AB_bc.csv and tu_sample_AB_bc.csv

These datasets have the same format as for tu_sample_bc.csv and tu_sample_he.csv, described above.

Auxiliary datasets

Time series plots: nation-wide sample

The Data folder also contains two datasets for generating aggregate time-series in Figure 1. The files tu_agg_bc.csv and tu_agg_heloc.csv contain time series of aggregate statistics throughout the sample.

These files both contain the following variables.

1. Run_Date is a date variable of the form 'YYYY-MM-01', indicating the month in which the data were reported by the bureau.
2. bal_avg is the average balance held by consumers during the month.
3. bal_sd is the standard deviation of balances held by consumers during the month.
4. bal_p25 is the lower quartile of balances held by consumers during the month.
5. bal_p50 is the median balance held by consumers during the month.
6. bal_p75 is the upper quartile of balances held by consumers during the month.

Time series plots: Alberta sample

The Data folder also contains another pair of datasets for generating aggregate time-series in Figure 8. The files tu_agg_AB_bc.csv and tu_agg_AB_heloc.csv contain time series of aggregate statistics throughout the sample, in an identical format, except that these were restricted to the province of Alberta.

Validation of aggregate credit-card balances

A dataset of time series of aggregate outstanding credit-card balances is required to generate Figure A1.1. These data are stored in a file TU_vs_BoC_totals.csv, which includes the following columns.

1. Date in DD/MM/YYYY format, representing the last day of each month.
2. MCP is a series drawn from the Webpage of the Bank of Canada, entitled Chartered bank selected assets: Month-end (formerly C1), Credit cards, which records assets in the personal loan category of non-mortgage loans.
3. tot_bal_all is the aggregate credit-card balance, in billions of Canadian dollars, across all institutions represented in the TransUnion database.
4. tot_bal_bank is the aggregate credit-card balance, in billions of Canadian dollars, across the chartered banks.

Validation of credit-card data coverage by province

A dataset of aggregate counts of the number of cardholders by province was compared to the population in each province in Figure A1.2. This information was collected in the dataset CC_TU_vs_StatsCan.csv, with the following columns.

1. region is the two-letter abbreviation of each province in Canada.
2. N_geq20_BC is the number of consumers aged 20 and above holding accounts during the month of January 2016.
3. geq20 is the population of each province in the age categories 20 and above, which was obtained from Statstics Canada Table: 17-10-0134-01, described below.

Other data

Other data were obtained to produce tables and figures of aggregate information about the credit-card and HELOC markets. These include:

• The Nilson Report, April 2020, Issue 1173, HSN Consultants, Inc.
• Chartered bank selected assets: Month-end (formerly C1), Credit cards, Bank of Canada, accessed June 2020.
• Estimates of population (2016 Census and administrative data), by age group and sex for July 1st, Canada, provinces, territories, health regions (2018 boundaries) and peer groups, Table: 17-10-0134-01, Statistics Canada, accessed June 2020.

These data sources are used in the statistical analysis that follows.

Statistical Analysis

These procedures were performed on a microcomputer to generate the tables and figures in the paper. These scripts are stored in the Code/Stats folder.

All Files in One Script:

1. Place all datasets in the Data folder, including the main datasets tu_sample_bc.csv, tu_sample_heloc.csv, tu_sample_AB_bc.csv, and tu_sample_AB_heloc.csv, along with the auxiliary datasets for time-series plots tu_agg_bc.csv, tu_agg_heloc.csv, tu_agg_AB_bc.csv, and tu_agg_AB_heloc.csv, and for figures in the appendix TU_vs_BoC_num_accts.csv and CC_TU_vs_StatsCan.csv.

2. Run COVID_CJE.sh in a terminal window from the Credit_COVID19_Canada folder.

This shell script calls the main R programs COVID_CJE_Cards.R, COVID_CJE_HELOCs.R COVID_CJE_AB_Cards.R, COVID_CJE_AB_HELOCs.R as well as the auxiliary R scripts CC_HE_time_series_figs.R, CC_BoC_vs_TU_comp_figs.R, and CC_TU_vs_StatsCan_comp_fig.R, all found in the Code/Stats folder, which analyze the datasets stored in the Data folder. These scripts create the tables and figures for the entire manuscript, by writing tex files to the Tables folder and eps files to the Figures folder.

Generating Sets of Files Separately

Nation-Wide Sample of Credit-Card Accounts

1. Place the dataset tu_sample_bc.csv in the Data folder.

2. Run Rscript COVID_CJE_Cards.R in a terminal window from the Credit_COVID19_Canada folder.

3. Obtain the tex files CC_KLD_vs_sample_01.tex and CC_KLD_kstep_fixed_vs_monthly_02.tex with numbers for columns 2 and 3 of Tables 1 and 2 from the Tables folder.

4. Obtain the images for panels (a) of Figures 2 and 3 in the eps files CC_hist_grp.eps and CC_3D_probs_discrete_1.eps from the Figures folder.

5. Obtain the images for Figures 4 and 6 in the eps files CC_sample_dev_pct_2020_MM.eps and CC_obs_vs_for_dev_pct_monthly_2020-MM.eps from the Figures folder, where MM represents the two-digit month of the Run_date after the close of the corresponding statement month.

Nation-Wide Sample of HELOC Accounts

1. Place the dataset tu_sample_heloc.csv in the Data folder.

2. Run Rscript COVID_CJE_HELOCs.R in a terminal window from the Credit_COVID19_Canada folder.

3. Obtain the tex files HE_KLD_vs_sample_01.tex and HE_KLD_kstep_fixed_vs_monthly_02.tex with numbers for columns 4 and 5 of Tables 1 and 2 from the Tables folder.

4. Obtain the images for panels (b) of Figures 2 and 3 in the eps files HE_hist_grp.eps and HE_3D_probs_discrete_1.eps from the Figures folder.

5. Obtain the images for Figures 5 and 7 in the eps files HE_sample_dev_pct_2020_MM.eps and HE_obs_vs_for_dev_pct_monthly_2020-MM.eps from the Figures folder, where MM represents the two-digit month of the Run_date after the close of the corresponding statement month.

Alberta Sample of Credit-Card Accounts

1. Place the dataset tu_sample_AB_bc.csv in the Data folder.
2. Run Rscript COVID_CJE_AB_Cards.R in a terminal window from the Credit_COVID19_Canada folder.
3. Obtain the tex file AB_CC_KLD_kstep_fixed_vs_monthly_02.tex with numbers for columns 2 and 3 for Table 3 in the Tables folder and panel (a) of Figure 9 in the file AB_CC_obs_vs_for_dev_pct_monthly_2015-11.eps in the Figures folder.

Alberta Sample of HELOC Accounts

1. Place the dataset tu_sample_AB_heloc.csv in the Data folder.
2. Run Rscript COVID_CJE_AB_HELOCs.R in a terminal window from the Credit_COVID19_Canada folder.
3. Obtain the tex file AB_HE_KLD_kstep_fixed_vs_monthly_02.tex with numbers for columns 4 and 5 for Table 3 in the Tables folder and panel (b) of Figure 9 in the file AB_HE_obs_vs_for_dev_pct_monthly_2015-11.eps in the Figures folder.

Auxilliary Tables and Figures

Instructions for generating the remaining tables and figures are outlined in the next section "Generating Tables and Figures Separately".

Generating Tables and Figures Separately

Tables

Table 1: Divergence from Sample Histograms

This Table contains information from two different modeling exercises: one for credit-cards and one for HELOCs.

For credit cards, run script COVID_CJE_Cards.R, which then runs script COVID_CJE_Cards_estim.R. Lines 118 to 199 of COVID_CJE_Cards_estim.R generate a file named CC_KLD_vs_sample_01.tex.

For HELOCs, run script COVID_CJE_HELOCs.R, which then runs script COVID_CJE_HELOCs_estim.R. Lines 118 to 199 of COVID_CJE_HELOCs_estim.R generate a file named HE_KLD_vs_sample_01.tex.

The numbers from these two tables are combined into the file Table_1.tex.

Table 2: Divergence from l-Step-Ahead Forecasts

This Table also contains information from two different modeling exercises: one for credit-cards and one for HELOCs.

For credit cards, run script COVID_CJE_Cards.R, which then runs script COVID_CJE_Cards_estim.R. Lines 363 to 433 of COVID_CJE_Cards_estim.R generate a file named CC_KLD_kstep_fixed_vs_monthly_02.tex. The columns under the

For HELOCs, run script COVID_CJE_HELOCs.R, which then runs script COVID_CJE_HELOCs_estim.R. Lines 363 to 433 of COVID_CJE_HELOCs_estim.R generate a file named HE_KLD_kstep_fixed_vs_monthly_02.tex.

The numbers from these two tables corresponding to the model with monthly transition matrices are combined into the file Table_2.tex.

Table 3: Divergence from l-Step-Ahead Forecasts in Alberta, 2015

The creation of this Table mirrors that of Table 2 on the Canadian population during the pandemic, except that it is run on a dataset restricted to consumers in the province of Alberta during the oil price shock in 2015. As with Table 2, it also contains information from two different modeling exercises: one for credit-cards and one for HELOCs.

For credit cards, run script COVID_CJE_Cards.R, which then runs script COVID_CJE_AB_Cards_estim.R. Lines 366 to 436 of COVID_CJE_AB_Cards_estim.R generate a file named AB_CC_KLD_kstep_fixed_vs_monthly_02.tex.

For HELOCs, run script COVID_CJE_HELOCs.R, which then runs script COVID_CJE_AB_HELOCs_estim.R. Lines 366 to 436 of COVID_CJE_AB_HELOCs_estim.R generate a file named AB_HE_KLD_kstep_fixed_vs_monthly_02.tex.

The numbers from these two tables corresponding to the model with monthly transition matrices are combined into the file Table_3.tex.

Table A1: Comparison of Accounts at the Credit Agency with Nation-Wide Totals in The Nilson Report

The set of numbers in the three leftmost columns are taken directly from the The Nilson Report, Issue 1173, April 2020, and are available here.

To compare with the contents of the TransUnion® database, we calculated the same summary statistics using the sample drawn from the database. The remaining information was obtained from running the script TU_vs_Nilson_comp.py, which produced the summary dataset called TU_vs_BoC_num_accts.csv, found in the Data folder.

Figures

Figure 1: Consumers' Outstanding Balances, 2017-2020

For credit cards, in panel (a), run script CC_HE_time_series_figs.R. Lines 91 to 132 generate a file named CC_time_series.eps from the data in a file named tu_agg_bc.csv.

For HELOCs, in panel (b), run script CC_HE_time_series_figs.R. Lines 141 to 182 generate a file named HE_time_series.eps from the data in a file named tu_agg_heloc.csv.

Figure 2: Histograms of Individuals' Balances

For credit cards, in panel (a), run script COVID_CJE_Cards.R, which then runs script COVID_CJE_Cards_prelim.R. Lines 53 to 71 of COVID_CJE_Cards_prelim.R generate a file named CC_hist_grp_sample.eps.

For HELOCs, in panel (b), run script COVID_CJE_HELOCs.R, which then runs script COVID_CJE_HELOCs_prelim.R. Lines 53 to 71 of COVID_CJE_HELOCs_prelim.R generate a file named HE_hist_grp_sample.eps.

Figure 3: Conditional Histograms of Individuals' Balances

For credit cards, in panel (a), run script COVID_CJE_Cards.R, which then runs script COVID_CJE_Cards_prelim.R. Lines 94 to 177 of COVID_CJE_Cards_prelim.R generate a file named CC_3D_probs_discrete_1.eps.

For HELOCs, in panel (b), run script COVID_CJE_HELOCs.R, which then runs script COVID_CJE_HELOCs_prelim.R. Lines 94 to 177 of COVID_CJE_HELOCs_prelim.R generate a file named HE_3D_probs_discrete_1.eps.

Figure 4: Deviations from Histograms (Credit Cards)

For credit cards, in panel (a), run script COVID_CJE_Cards.R, which then runs script COVID_CJE_Cards_estim.R. Lines 36 to 114 of COVID_CJE_Cards_estim.R generate a set of files named CC_sample_dev_pct_2020_MM.eps.

Figure 5: Deviations from Histograms (HELOCs)

For HELOCs, in panel (b), run script COVID_CJE_HELOCs.R, which then runs script COVID_CJE_HELOCs_estim.R. Lines 36 to 114 of COVID_CJE_HELOCs_estim.R generate a set of files named HE_sample_dev_pct_2020_MM.eps.

Figure 6: Deviations from Forecasted Credit-Card Balances

For credit cards, in panel (a), run script COVID_CJE_Cards.R, which then runs script COVID_CJE_Cards_estim.R. Lines 473 to 506 of COVID_CJE_Cards_estim.R generate a file named CC_obs_vs_for_dev_pct_monthly_2020_MM.eps.

Figure 7: Deviations from Forecasted HELOC Balances

For HELOCs, in panel (b), run script COVID_CJE_HELOCs.R, which then runs script COVID_CJE_HELOCs_estim.R. Lines 473 to 506 of COVID_CJE_HELOCs_estim.R generate a file named HE_obs_vs_for_dev_pct_monthly_2020_MM.eps.

Figure 8: Consumers' Outstanding Balances, Alberta, 2012-2016

For credit cards, in panel (a), run script CC_HE_time_series_figs.R. Lines 226 to 267 generate a file named AB_CC_time_series.eps from the data in a file named tu_agg_AB_bc.csv.

For HELOCs, in panel (b), run script CC_HE_time_series_figs.R. Lines 276 to 317 generate a file named AB_HE_time_series.eps from the data in a file named tu_agg_AB_heloc.csv.

Figure 9: Deviations from Forecasted Balances in Alberta, October 2015

For credit cards, in panel (a), run script COVID_CJE_AB_Cards.R, which then runs script COVID_CJE_AB_Cards_estim.R. Lines 476 to 509 of COVID_CJE_AB_Cards_estim.R generate a file named AB_CC_obs_vs_for_dev_pct_monthly_2015-11.eps.

For HELOCs, in panel (b), run script COVID_CJE_AB_HELOCs.R , which then runs script COVID_CJE_AB_HELOCs_estim.R. Lines 476 to 509 of COVID_CJE_AB_HELOCs_estim.R generate a file named AB_HE_obs_vs_for_dev_pct_monthly_2015-11.eps.

Figure A1.1: Time Series of Aggregate Credit-Card Balances

The two panels, which are both generated with the same script, one showing balances and the other showing percent changes of all the series. Two of the series were created using the sample from the TransUnion database with the following script: TU_vs_BoC_comp.py

The other series is derived from an internal database housed at the Bank of Canada and collected from regulatory returns. It is available on the "Banking and Financial Statistics" Webpage of the Bank of Canada and is called Chartered bank selected assets: Month-end (formerly C1). We use the row of the table labeled "Credit cards".

Together, the aggregate time-series data are recorded in the file TU_vs_BoC_totals.csv. The figures in the file TU_vs_BoC_comparison.eps are then generated with the script CC_TU_vs_BoC_comp_figs.R, on lines 89 to 140.

Figure A1.2: Credit Data Coverage for Adults in Canada, by Province

The numbers in this figure were calculated with the scripts TU_vs_StatsCan_comp.py and CC_TU_vs_StatsCan_comp_fig.R in the Code folder. It requires the dataset CC_TU_vs_StatsCan.csv, comprising the number of credit-card account holders aged 20 and above and the figures obtained from Statistics Canada in the table called *Estimates of population (2016 Census and administrative data), by age group and sex for July 1st, Canada, provinces, territories, health regions (2018 boundaries) and peer groups, * Table: 17-10-0134-01. The file CC_TU_vs_StatsCan_comp.eps for Figure A1.2 is created by running lines 96 to 113 of the script CC_TU_vs_StatsCan_comp_fig.R.

Computing Requirements

Data Manipulation

The csv files in the Data folder were generated on the EDITH 2.0 High Performance cluster housed at the Bank of Canada.

For the queries that generated the datasets, 36 CPUs with 240 GB of memory were sufficient to create the datasets within at most 24 hours each.

Statistical Analysis

Once the datasets have been saved in the Data folder, the remaining analysis, including the generation of all the tables and figures in the paper can be performed on a single microcomputer, such as a laptop computer. The particular model of computer on which the statistical analysis was run is a Dell Precision 3520, running a 64-bit Windows 10 operating system, with a 4-core x64-based processor, model Intel(R) Core(TM) i7-7820HQ CPU, running at 2.90GHz, with 16 GB of RAM.

Software

Data Manipulation

The data manipulation was conducted using a NoSQL dialect called Apache Spark, which is based on the functional programming language Scala and was implemented with PySpark in Python. The scripts were run using the Anaconda 2 distribution, version 4.3.1, with Python version 2.7 and PySpark version 2.3.0.

The batch jobs were submitted to the computing cluster using batch scheduling software called SLURM.

Other resources used to run the batch jobs include:

• sbt, version 1.3.6, which is a build tool for Scala, Java, among others
• java, version 1.8.0_141

Statistical Analysis

The statistical analysis was conducted in R, version 4.0.2, which was released on June 22, 2020, on a 64-bit Windows platform x86_64-w64-mingw32/x64.

The attached packages include the following:

• data.table, version 1.13.0 (using 4 threads), to handle the main data table for analysis in the _prelim.R and _estim.R scripts.

• xtable, version 1.8-4, to generate LaTeX tables for Tables 1, 2, and 3.

• plot3D, version 1.3, to produce a 3-D bar chart of transition frequency, which created the plots in Figure 3.

• MASS, version 7.3-51.6, was also used to estimate the smoothed surface of the transition density as an alternative to that in Figure 3 but was not included in the paper.

The creation of other figures, including Figures A1.1 and A1.2, required the following packages for data manipulation and graphics:

• openxlsx, version 4.2.3
• dplyr, version 1.0.5
• lubridate, version 1.7.10
• ggplot2, version 3.3.3
• ggpubr, version 0.4.0
• ggthemes, version 4.2.4
• Cairo, version 1.5-12.2

Upon attachment of the above packages, the following packages were loaded via a namespace, but not attached, with the following versions:

• Rcpp version 1.0.5
• lattice version 0.20-41
• grid version 4.0.2
• DTMCPack version 0.1-2
• stats4 version 4.0.2
• magrittr version 1.5
• RcppParallel version 5.0.2
• misc3d version 0.8-4
• markovchain version 0.8.5-3
• Matrix version 1.2-18
• tools version 4.0.2
• igraph version 1.2.6
• parallel version 4.0.2
• compiler version 4.0.2
• pkgconfig version 2.0.3
• matlab version 1.0.2
• nnet version 7.3-14
• expm version 0.999-5
• zip version 2.1.1
• cellranger version 1.1.0
• pillar version 1.6.0
• forcats version 0.5.1
• lifecycle version 1.0.0
• tibble version 3.1.0
• gtable version 0.3.0
• rlang version 0.4.10
• curl version 4.3
• haven version 2.3.1
• rio version 0.5.26
• stringr version 1.4.0
• withr version 2.4.2
• hms version 1.0.0
• generics version 0.1.0
• vctrs version 0.3.7
• grid version 4.0.5
• tidyselect version 1.1.0
• glue version 1.4.2
• R6 version 2.5.0
• rstatix version 0.7.0
• fansi version 0.4.2
• readxl version 1.3.1
• foreign version 0.8-81
• carData version 3.0-4
• purrr version 0.3.4
• tidyr version 1.1.3
• car version 3.0-10
• scales version 1.1.1
• backports version 1.2.1
• ellipsis version 0.3.1
• abind version 1.4-5
• colorspace version 2.0-0
• ggsignif version 0.6.1
• utf8 version 1.2.1
• stringi version 1.5.3
• munsell version 0.5.0
• broom version 0.7.6
• crayon version 1.4.1

References

• Trade-Level Database, Run Dates 2017-01-01 to 2020-09-01, TransUnion, accessed October 2020.
• Consumer Risk Characteristics Database, Run Dates 2017-01-01 to 2020-09-01, TransUnion, accessed October 2020.
• Estimates of population (2016 Census and administrative data), by age group and sex for July 1st, Canada, provinces, territories, health regions (2018 boundaries) and peer groups, Table: 17-10-0134-01, Statistics Canada, accessed June 2020.
• The Nilson Report, April 2020, Issue 1173, HSN Consultants, Inc., url: https://nilsonreport.com/publication_newsletter_archive_issue.php?issue=1173
• Chartered bank selected assets: Month-end (formerly C1), Credit cards, Bank of Canada, accessed June 2020.

Acknowledgements

The views expressed are those of the authors; no responsibility for these views should be attributed to the Bank of Canada; all errors are the responsibility of the authors. We thank the HPC team at the Bank of Canada for their excellent assistance with the EDITH 2.0 High Performance Cluster.