We maintain an allowance for loan losses for HFI loans held by Fannie Mae and by consolidated Fannie Mae MBS trusts, excluding loans for which we have elected the fair value option. When calculating our allowance for loan losses, we consider the unpaid principal balance, net of unamortized premiums and discounts, and other cost basis adjustments of HFI loans at the balance sheet date. We record write-offs as a reduction to our allowance for loan losses at the point of foreclosure, completion of a short sale, upon the redesignation of nonperforming and reperforming loans from HFI to HFS or when a loan is determined to be uncollectible.

The following table displays changes in our allowance for single-family loans, multifamily loans and total allowance for loan losses, including the transition impact of adopting Accounting Standards Update (“ASU”) 2016-13, “Financial Instruments—Credit Losses (Topic 326), Measurement of Credit Losses on Financial Instruments,” which was later amended by ASU 2019-04, ASU 2019-05 and ASU 2019-1 (collectively, the “CECL standard”), on January 1, 2020. See “Note 1, Summary of Significant Accounting Policies” in our 2020 Form 10-K for additional information on changes to our allowance for loan losses resulting from our adoption of the CECL standard.

The benefit or provision for loan losses excludes provision for accrued interest receivable losses, guaranty loss reserves and credit losses on available-for-sale (“AFS”) debt securities. Cumulatively, these amounts are recognized as “Benefit (provision) for credit losses” in our condensed consolidated statements of operations and comprehensive income.


	For the Three Months Ended September 30,				For the Nine Months Ended September 30,
	2021		2020		2021		2020
	(Dollars in millions)
Single-family allowance for loan losses:
Beginning balance	$	(6,064)	$	(11,598)	$	(9,344)	$	(8,759)
Transition impact of the adoption of the CECL standard	—		—		—		(1,229)
Benefit (provision) for loan losses	839		931		4,003		(857)
Write-offs	51		261		271		377
Recoveries	(133)		(4)		(212)		(10)
Other	(8)		23		(33)		91
Ending Balance	$	(5,315)	$	(10,387)	$	(5,315)	$	(10,387)
Multifamily allowance for loan losses:
Beginning balance	$	(1,050)	$	(1,368)	$	(1,208)	$	(257)
Transition impact of the adoption of the CECL standard	—		—		—		(493)
Benefit (provision) for loan losses	45		(34)		175		(669)
Write-offs	7		86		54		104
Recoveries	(21)		—		(40)		(1)

Ending Balance	$	(1,019)	$	(1,316)	$	(1,019)	$	(1,316)
Total allowance for loan losses:
Beginning balance	$	(7,114)	$	(12,966)	$	(10,552)	$	(9,016)
Transition impact of the adoption of the CECL standard	—		—		—		(1,722)
Benefit (provision) for loan losses	884		897		4,178		(1,526)
Write-offs	58		347		325		481
Recoveries	(154)		(4)		(252)		(11)
Other	(8)		23		(33)		91
Ending Balance	$	(6,334)	$	(11,703)	$	(6,334)	$	(11,703)

Our benefit or provision for loan losses can vary substantially from period to period based on a number of factors, such as changes in actual and forecasted home prices or property valuations, fluctuations in actual and forecasted interest rates, borrower payment behavior, events such as natural disasters or pandemics, the type, volume and effectiveness of our loss mitigation activities, including forbearances and loan modifications, the volume of foreclosures completed, and the redesignation of loans from HFI to HFS. Our benefit or provision can also be impacted by updates to the models, assumptions, and data used in determining our allowance for loan losses. As described below, our benefit or provision for loan losses and our loss reserves have been significantly affected by our estimates of the impact of the COVID-19 pandemic and the pace and strength of the economy’s subsequent recovery, which require significant management judgment.

•Benefit from actual and forecasted home price growth. For the three months ended September 30, 2021, actual home price growth continued to be strong. We also increased our expectations for home price growth on a national basis for full-year 2021. Higher home prices decrease the likelihood that loans will default and reduce the amount of loan loss on loans that do default, which impacts our estimate of losses and ultimately reduces our loss reserves and provision for loan losses.

•Benefit from the redesignation of certain nonperforming and reperforming single-family loans from HFI to HFS. We redesignated certain nonperforming and reperforming single-family loans from HFI to HFS, as we no longer intend to hold them for the foreseeable future or to maturity. Upon redesignation of these loans, we recorded the loans at the lower of cost or fair value with a write-off against the allowance for loan losses. Amounts recorded in the allowance related to these loans exceeded the amounts written off, resulting in a net benefit for loan losses.

•Benefit from changes in assumptions regarding COVID-19 forbearance and loan delinquencies. During the first and second quarters of 2021, management used its judgment to supplement the loss projections developed by our credit loss model to account for uncertainty arising from the COVID-19 pandemic that was not represented in historical data or otherwise captured by our credit model. As of September 30, 2021, management removed the remaining non-modeled adjustment as the effects of the economic stimulus, the vaccine rollout, and the effectiveness of COVID-19-related loss mitigation strategies have all become much less uncertain. Specifically, the decrease in uncertainty as of September 30, 2021 compared with the end of 2020 was primarily driven by the passage of the American Rescue Plan Act of 2021 and the broad implementation of the COVID-19 vaccination program in the United States, which contributed to a significant increase in business activity and helped support continued economic growth. There has also been a steady decline in the number of borrowers in a COVID-19-related forbearance, lessening expectations of loan losses. Additionally, we believe the array of possible future economic environments included in our credit model, which captures scenarios that may be remote, combined with data consumed over the course of the COVID-19 pandemic have removed the need to continue to supplement modeled results.

•Provision for higher actual and projected interest rates. Actual and projected interest rates were higher as of September 30, 2021 compared with December 31, 2020. As mortgage rates increase, we expect a decrease in future prepayments on single-family loans, including modified loans. Lower expected prepayments extend the expected lives of modified loans, which increases the expected impairment relating to term and interest-rate concessions provided on these loans, resulting in a provision for loan losses.

The decrease in single-family write-offs for the three months ended September 30, 2021 compared with the three months ended September 30, 2020 was primarily due to a reduction in the number of nonperforming and reperforming loans redesignated from HFI to HFS in the third quarter of 2021 as well as more favorable pricing on our HFS loans.

•Provision for changes in assumptions regarding COVID-19 forbearance and loan delinquencies, which included adjustments to modeled results. Our single-family provision for loan losses for the nine months ended September 30, 2020 was driven by the economic dislocation caused by the COVID-19 pandemic, with the majority of the provision recognized in the first quarter of 2020. Estimating expected loan losses as a result of the COVID-19 pandemic continued to require significant management judgment regarding a number of matters, including our expectations surrounding borrower participation in a COVID-19-related forbearance, the type and extent of loss mitigation that may be needed when the loan exits forbearance, the high degree of uncertainty regarding the future course of the pandemic and its effect on the economy, and expectations regarding the impact of fiscal stimulus to support borrowers. As a result, the model used to estimate single-family loan losses did not capture the entirety of losses we expected to incur relating to COVID-19 at that time. The model consumed data from the initial months of the pandemic, including loan delinquencies, and updated credit profile data for loans in forbearance. As more of this data was consumed by our credit loss model, we reduced the non-modeled adjustment initially recorded in the first quarter.

In the third quarter of 2020, management continued to apply its judgment and supplement model results as of September 30, 2020, taking into account the continued high degree of uncertainty regarding the future impact of the pandemic and its effect on the economy, future economic and housing policy, and extended foreclosure moratoriums. These factors, combined with higher loan delinquencies, led to an increase in provision attributable to these COVID-19-related factors, which was partially offset by a decrease in our estimated single-family cumulative forbearance take-up, based on the economic data and forbearance activity observed through the third quarter of 2020.

The factors discussed above were offset by the factors below, which contributed to a single-family benefit for loan losses for the three months ended September 30, 2020 and reduced the amount of single-family provision for loan losses recognized for the nine months ended September 30, 2020:

•Benefit from lower actual and projected interest rates. For much of 2020, we continued to be in a historically low interest-rate environment. As mortgage interest rates decline, we expect an increase in future prepayments on single-family loans, including modified loans. Higher expected prepayments shorten the expected lives of modified loans, which decreases the expected impairment relating to term and interest-rate concessions provided on these loans and results in a benefit for loan losses. Most of this benefit from lower actual and projected mortgage interest rates was recognized in the first half of 2020.

•Benefit from the redesignation of certain reperforming single-family loans from HFI to HFS. In the third quarter of 2020, we resumed sales of reperforming loans after our suspension of new loan sales in the second quarter of 2020. As a result, we redesignated certain reperforming single-family loans from HFI to HFS in the third quarter of 2020, as we no longer intended to hold them for the foreseeable future or to maturity. Upon redesignation of these loans, we recorded the loans at the lower of cost or fair value with a write-off against the allowance for loan losses. Amounts recorded in the allowance related to these loans exceeded the amounts written off, which contributed to a net benefit for loan losses for the three and nine months ended September 30, 2020.

•Benefit from actual and expected home price growth. In the first quarter of 2020, we significantly reduced our expectations for home price growth to near-zero for 2020. However, the negative impact from the first quarter of 2020 was more than offset by an increase in actual home price growth in the second and third quarters.

•Benefit from actual and projected economic data. For the three and nine months ended September 30, 2021, property value forecasts increased due to continued demand for multifamily housing. In addition, improved job growth led to an increase in projected average property net operating income, which reduced the probability of loan defaults resulting in a benefit for loan losses for the quarter and year-to-date.

•Benefit from lower expected loan losses as a result of the COVID-19 pandemic. Similar to our single-family benefit for loan losses described above, for the first and second quarters of 2021 management used its judgment to supplement the loss projections developed by our credit loss model to account for uncertainty arising from the COVID-19 pandemic. As of September 30, 2021, management removed the remaining non-modeled adjustment as the effects of the economic stimulus, the vaccine rollout, and the effectiveness of COVID-19-related loss mitigation strategies have all become much less uncertain. See our single-family 2021 benefit for loan losses discussion above for more information on these factors.

Our multifamily provision for loan losses for the nine months ended September 30, 2020 was driven by higher expected losses as a result of the economic dislocation caused by the COVID-19 pandemic and heightened economic uncertainty, driven by elevated unemployment which we expected would result in a decrease of multifamily property income and property values. In addition, the multifamily provision for loan losses included increased expected loan losses on seniors housing loans as these properties were disproportionately impacted by the pandemic. The vast majority of these expenses were recognized in the first half of 2020. Consistent with the single-family discussion above, the model we used to estimate multifamily loan losses did not capture the entirety of losses we expected to incur relating to COVID-19 at that time. The model consumed data from the initial months of the pandemic, but we continued to apply management judgment and supplement model results as of September 30, 2020, taking into account the continued high degree of uncertainty that remained relating to the impact of the pandemic.

In the third quarter of 2020, our provision for expected multifamily loan losses as a result of the COVID-19 pandemic was relatively flat. In September 2020, we decreased our estimate for loan losses due to a downward revision of our estimated multifamily cumulative forbearance take-up rate, as well as an improved forecasted unemployment rate. These benefits were offset by continued economic uncertainty, forbearance arrangements that were extended beyond the initial term, and overall increased delinquencies. These factors, inclusive of the other components of the multifamily provision for loan losses, resulted in a modest expense for the third quarter of 2020.