KEYCORP /NEW/

The allowance for loan and lease losses represents management’s estimate of all expected credit losses over the expected contractual life of our existing loan portfolio. Determining the appropriateness of the allowance is complex and requires judgment by management about the effect of matters that are inherently uncertain. These critical estimates include significant use of our own historical data and complex methods to interpret them. We have an ongoing process to evaluate and enhance the quality, quantity, and timeliness of our data and interpretation methods used in the determination of these allowances. These evaluations are inherently subjective, as they require material estimates and may be susceptible to significant change, and include, among others:

•PD,

•LGD,

•Outstanding balance of the loan,

•Movement through delinquency stages,

•Amounts and timing of expected future cash flows,

•Value of collateral, which may be obtained from third parties,

•Economic forecasts which are obtained from a third party provider, and

•Qualitative factors, such as changes in current economic conditions, that may not be reflected in modeled results.

As described in our accounting policy related to the ALLL in Note 1 (“Summary of Significant Accounting Policies”) of this report under the heading “Allowance for Loan and Lease Losses," we employ a disciplined process and methodology to establish our ALLL, which has three main components: (i) asset specific / individual loan reserves; (ii) quantitative (formulaic or pooled) reserves; and (iii) qualitative (judgmental) reserves.

We use a non-DCF factor-based approach to estimate expected credit losses that include component PD/LGD/EAD models as well as less complex estimation methods for smaller loan portfolios. Probability of default models estimate the likelihood a borrower will cease making payments as agreed. These models use observed loan-level information and projected paths of macroeconomic variables. Borrower credit attributes including FICO scores of consumers and internally assigned risk ratings for commercial borrowers are significant inputs to the models. Consumer FICO scores are refreshed quarterly and commercial risk ratings are updated annually with select borrowers updated more frequently. The macroeconomic trends that have a significant impact on the probability of default vary by portfolio segment. Exposure at default models estimate the loan balance at the time of default. We use an amortization based formulaic approach to estimate account level EAD for all term loans. We use portfolio specific methods in each of our revolving product portfolios. LGD models estimate the loss we will suffer once a loan is in default. Account level inputs to LGD models include collateral attributes, such as loan to value.

If we observe limitations in the data or models, we use model overlays to make adjustments to model outputs to capture a particular risk or compensate for a known limitation. These variables and others may result in actual loan losses that differ from the originally estimated amounts.

This estimate produced by our models is forward-looking and requires management to use forecasts about future economic conditions to determine the expected credit loss over the remaining life of an instrument. Moody’s Consensus forecast is our source of macroeconomic projections, including the interest rate forecasts used in the credit models. We use a two year reasonable and supportable period across all products to forecast economic conditions. As the length of the life of a financial asset increases, these inputs may become impractical to estimate as reasonable and supportable. We believe the two year time horizon appropriately aligns with our business planning, available industry guidance, and reliability of various forecasting services. Following this two year period in which supportable forecasts can be generated, for all modeled loan portfolios, we revert expected credit losses to a level that is consistent with our historical information by reverting the macroeconomic variables (model inputs) to their long run average. We revert to historical loss rates for less complex estimation methods for smaller portfolios.

A four quarter reversion period is used where the macroeconomic variables linearly revert to their long run average following the two year reasonable and supportable period. We use a 20 year lookback period for determining long run historical average of the macroeconomic variables. We determined the 20 year lookback period is appropriate as it captures the previous two economic cycles as well as the impact from the pandemic.

The ALLL is sensitive to various macroeconomic drivers such as GDP and unemployment as well as portfolio attributes such as remaining term, outstanding balance, risk ratings, FICO, LTV, and delinquency status. Our ALLL models were designed to capture the correlation between economic and portfolio changes. As such, evaluating shifts in individual portfolio attributes and macroeconomic variables in isolation may not be indicative of past or future performance.

It is difficult to estimate how potential changes in any one factor or input might affect the overall ALLL because we consider a wide variety of factors and inputs in estimating the ALLL. Changes in the factors and inputs considered may not occur at the same rate and may not be consistent across all geographies or product types, and changes in factors and input may be directionally inconsistent, such that improvement in one factor may offset deterioration in others. However, to consider the impact of a hypothetical alternate economic forecast, we compare the modeled quantitative allowance results using a downside economic scenario. The maximum difference in the quarterly macroeconomic variables as of December 31, 2025, between the base and downside scenarios over the two year reasonable and supportable period includes an approximate 6 percentage point decline in GDP annualized growth and an approximate 4 percentage point increase in the U.S. unemployment rate. The difference between these two scenarios would have driven an increase of approximately 1.7x for commercial and 1.6x for the consumer modeled allowance results.

Similarly, deteriorating conditions for portfolio factors were also considered by moderately stressing key portfolio drivers, relative to the baseline portfolio conditions. Stressing risk ratings by two ratings for commercial loans generates a 1.5x increase in the commercial modeled allowance results. Stressing FICO by ten points, and LTV and utilization by 10% for consumer loans generates a 1.1x increase in the consumer modeled allowance results.

Note that these analyses demonstrate the sensitivity of the ALLL to key quantitative assumptions, but exclude potential impacts to non-modeled allowance components and reserves for individually assessed loans. Furthermore, these analyses are not intended to estimate changes in the overall ALLL as they do not reflect qualitative factors, changes in current economic conditions that may not be reflected in quantitatively derived results, and other relevant factors that must be considered to ensure the ALLL reflects our best estimate of current expected credit losses.