Reserves for credit losses represent our best estimate of the expected credit losses in our outstanding portfolio of Card Member loans and receivables as of the balance sheet date. The CECL methodology, which became effective January 1, 2020, requires us to estimate lifetime expected credit losses by incorporating historical loss experience, as well as current and future economic conditions over a reasonable and supportable period (R&S Period), which is approximately three years, beyond the balance sheet date. We make various judgments combined with historical loss experience to determine a reserve rate that is applied to the outstanding loan or receivable balance to produce a reserve for expected credit losses.

We use a combination of statistically-based models that incorporate current and future economic conditions throughout the R&S Period. The process of estimating expected credit losses is based on several key models: Probability of Default (PD), Exposure at Default (EAD), and future recoveries for each month of the R&S Period. Beyond the R&S Period, we estimate expected credit losses by immediately reverting to long-term average loss rates.

•EAD models are used to estimate the balance of an account at the time of write-off. This includes balances less expected repayments based on historical payment and revolve behavior, which vary by customer. Due to the nature of revolving loan portfolios, the EAD models are complex and involve assumptions regarding the relationship between future spend and payment behaviors.

•Recovery models are used to estimate amounts that are expected to be received from Card Members after default occurs, typically as a result of collection efforts. Future recoveries are estimated taking into consideration the time of default, time elapsed since default and macroeconomic conditions.

We also estimate the likelihood and magnitude of recovery of previously written off accounts considering how long ago the account was written off and future economic conditions. Our models are developed using historical loss experience covering the economic cycle and consider the impact of account characteristics on expected losses.

Future economic conditions that are incorporated over the R&S Period include multiple macroeconomic scenarios provided to us by an independent third party. Management reviews these economic scenarios and applies judgment to weight them in order to reflect the uncertainty surrounding these scenarios. These macroeconomic scenarios contain certain variables, including unemployment rates and real gross domestic product, that are significant to our models.

We also evaluate whether to include qualitative reserves to cover losses that are expected but, in our assessment, may not be adequately represented in the quantitative methods or the economic assumptions. We consider whether to adjust the quantitative reserves (higher or lower) to address possible limitations within the models or factors not included within the models, such as external conditions, emerging portfolio trends, the nature and size of the portfolio, portfolio concentrations, the volume and severity of past due accounts, or management risk actions.

Lifetime losses for most of our loans and receivables are evaluated at an appropriate level of granularity, including assessment on a pooled basis where financial assets share similar risk characteristics, such as past spend and remittance behaviors, credit bureau scores where available, delinquency status, tenure of balance outstanding, amongst others. Credit losses on accrued interest are measured and presented as part of Reserves for credit losses on the Consolidated Balance Sheets and within the Provisions for credit losses in the Consolidated Statements of Income, rather than reversing interest income. Separate models are used for accounts deemed a troubled debt restructuring, which are measured individually using a discounted cash flow model. See Note 2 for information on troubled debt restructurings.

Loans and receivable balances are written off when we consider amounts to be uncollectible, which is generally determined by the number of days past due and is typically no later than 180 days past due for pay in full or revolving loans and 120 days past due for term loans. Loans and receivables in bankruptcy or owed by deceased individuals are generally written off upon notification.

Results for reporting periods beginning January 1, 2020 are presented using the CECL methodology while comparative information continues to be reported in accordance with the incurred loss methodology in effect for prior years. Reserves for credit losses under the incurred loss methodology were primarily based upon statistical and analytical models that analyzed portfolio performance and reflected management’s judgments regarding the quantitative components of the reserve. The models considered several factors, including delinquency-based loss migration rates, loss emergence periods and average losses and recoveries over an appropriate historical period. Similar to the CECL methodology, we considered whether to adjust the quantitative reserves for certain external and internal qualitative factors, which may increase or decrease the reserves for credit losses.

Card Member loans reserve for credit losses increased for the year ended December 31, 2020, primarily driven by deterioration of the global macroeconomic outlook, including unemployment and GDP, partially offset by improved credit performance and a decline in outstanding balances.


(Millions)	2020		2019		2018
Beginning Balance(a)	$	4,027	$	2,134	$	1,706
Provisions(b)	3,453		2,462		2,266
Net write-offs (c)
Principal	(1,795)		(1,860)		(1,539)
Interest and fees	(375)		(375)		(304)
Other(d)	34		22		5
Ending Balance	$	5,344	$	2,383	$	2,134

(c)Principal write-offs are presented less recoveries of $568 million, $525 million and $444 million for the years ended December 31, 2020, 2019 and 2018, respectively. Recoveries of interest and fees were not significant. Amounts include net (write-offs) recoveries from TDRs of $(134) million, $(79) million and $(33) million for the years ended December 31, 2020, 2019 and 2018, respectively.

(d)Primarily includes foreign currency translation adjustments of $35 million, $4 million and $(11) million for the years ended December 31, 2020, 2019 and 2018, respectively.

Card Member receivables reserve for credit losses increased for the year ended December 31, 2020, primarily driven by deterioration of the global macroeconomic outlook, including unemployment and GDP, partially offset by improved credit performance and a decline in outstanding balances.


(Millions)	2020		2019		2018
Beginning Balance(a)	$	126	$	573	$	521
Provisions(b)	1,015		963		937
Net write-offs(c)	(881)		(900)		(859)
Other(d)	7		(17)		(26)
Ending Balance	$	267	$	619	$	573

(c)Net write-offs are presented less recoveries of $386 million, $374 million and $367 million for the years ended December 31, 2020, 2019 and 2018, respectively. Amounts include net recoveries (write-offs) from TDRs of $(47) million, $(16) million and nil, for the years ended December 31, 2020, 2019 and 2018, respectively.