Time-Dependent PD and LGD in IFRS 9 Expected Credit Loss Models
Summary
The discussion asks whether expected credit loss calculations need probability of default (PD) and loss given default (LGD) estimates that vary with time in a staging category. It describes a common IFRS 9 approach: estimate through-the-cycle (TTC) parameters, including marginal PDs across an exposure’s life, then adjust them using point-in-time (PiT) models tied to macroeconomic forecasts. The forecast horizon is usually limited, while longer-term estimates remain anchored to the TTC component.
A second answer frames time dependence as a trade-off between model complexity and accuracy. It says LGD is often held constant while PD varies, though practice depends on the firm and purpose. Time in stage may matter most for stage 2, where an account can transition toward default; stage 3 accounts are already in default. The replies offer practice-oriented guidance, not a universal specification, and note that the appropriate detail depends on modeling objectives and applicable measurement requirements.
Key ideas
- IFRS 9 ECL models commonly combine through-the-cycle estimates with point-in-time macroeconomic adjustments.
- Marginal PD estimates can vary across the life of an exposure.
- Forecasts typically affect near-term estimates, while longer horizons rely more on the through-the-cycle component.
- Model time dependence should be weighed against its added complexity and accuracy.
- Stage 2 may be particularly sensitive to time spent in the stage, while stage 3 accounts are already in default.
Tags
Full text
# PD and LGD for ECL calculations needs to be time dependent?
# PD and LGD for ECL calculations needs to be time dependent?
I'm studying the implementation of an expected credit loss (ECL) model. I have encountered a complication. Do I need to calculate a probability of default (PD) and loss given default (LGD) with a time dependency of the staging position, i.e. for a specific portfolio the parameters needs to be: $PD_{t,i}$ and $LGD_{t,i}$, where $t$ is the time the account is on the respective staging and $i$ is the staging value (1,2 or 3). Or I can simply have $PD_{i}$ and $LGD_{i}$ only depending on the portfolio and the staging.
Thank you very much in advance.
## Answer by Richi Wa (score 3)
https://quant.stackexchange.com/a/51258
I assume that you calculate ECL in the context of IFRS9 -correct?
market practice often follows the following approach:
- estimate a TTC PD/LGD (TTC = through the cycle). This corresponds to your lifetime estimate (e.g. one marginal PD value for each year of the life of your exposure) in the average of the economic cycle.
- But for IFRS9 provisioning you have to reflect current information. Thus usually a PiT (point in time) model is developed on top of the TTC component. In this step you model how your PD/LGD estimate depends on macro economic variables. Later when you calculate provisions you perform (or just retrieve) a forecast of the macros and adjust your PD or LGD values according to this forecast and the sensitivity modeled in the PiT part. Usually forecasts are only used for ~2 years. Noone expects you to forecast further into the future.
In summary you have one time dimension in the sense of marginal PDs for each year (or even month) of the duration of your exposures. Another time dimension is how the forthcoming years adjust these estimates.
## Answer by Attack68 (score 2)
https://quant.stackexchange.com/a/51257
You are building a model - the question you are asking is a trade off between accuracy and complexity.
If the accuracy only improves in a minor capacity and the extension is considered complex you can ask the question "is it really necessary"?
If the accuracy greatly improves then whether the extension is considered complex or not I suspect that for whatever purpose it is probably more beneficial to include it.
Most of the time I have seen an LGD kept constant with the PD being flexible to vary over time, but models do vary across firms and purposes.
Time evolution in each staging area might be more relevant. If it is the IFRS stage definition then stage 3 (loans in default) are already in default so time is irrelevant, otherwise stage 2 is probably most sensitive to time in the stage, becuase they are in transition between stage 1 and stage 2 if a further default event occurs.
## Answer by user51037 (score 2)
https://quant.stackexchange.com/a/59585
To answer your question in the commentary above: IFRS 9 requires each financial instrument to be measured individuallyShown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.