Lifetime PD is the probability that a borrower defaults at any point between the reporting date and the end of the loan’s expected life. Under IFRS 9, it drives Stage 2 expected credit loss and is built as a term structure of annual or monthly PDs that reflects current conditions and forward-looking forecasts.
Every IFRS 9 impairment model needs a default probability for each future period of a loan. Together, those probabilities form a PD term structure. Lifetime PD is the cumulative result. However, many teams struggle to move from a 12-month PD model to a full term structure.
This guide explains lifetime PD in plain terms. First, you will see how it differs from 12-month PD. Next, you will learn the cumulative, marginal and conditional PD relationships, with a five-year worked table. Then we cover three ways to build the term structure, how to make it forward-looking, and how it feeds expected credit loss (ECL).
IFRS 9 sorts exposures into three stages, and the stage sets the PD horizon. First, Stage 1 exposures carry 12-month ECL. Second, exposures with a significant increase in credit risk (SICR) move to Stage 2 and carry lifetime ECL. Third, credit-impaired exposures sit in Stage 3, where lifetime ECL also applies. Ind AS 109 follows the same three-stage model, so this guide applies to Ind AS reporters too.
As a result, a loan that moves from Stage 1 to Stage 2 can see its provision jump sharply. The longer PD horizon drives most of that jump. Therefore, the quality of your term structure matters as much as the quality of your 12-month PD.
| Feature | 12-month PD | Lifetime PD |
| Horizon | Next 12 months from the reporting date (or to maturity, if shorter) | Remaining expected life of the exposure |
| Used for | Stage 1 ECL | Stage 2 ECL |
| Output | One PD per borrower or segment | A PD for each future period (the term structure) |
| Typical inputs | Rating or score, current conditions, one-year macro forecast | The same inputs, plus time on book, maturity and multi-year macro paths |
| Main risk | Missing a deterioration that emerges later | Compounding errors over long horizons |
Stage 3 exposures have already defaulted. Consequently, their PD is effectively 100%, and measurement focuses on LGD and exposure.
Many banks start from the PD models they built for Basel capital. That is a sensible base, and the Basel-style work is covered in our credit risk modeling resources. However, IFRS 9 asks for different properties.
Consequently, a Basel PD usually needs adjustment before it can feed an IFRS 9 model. IFRS 9 requires an unbiased, probability-weighted estimate that reflects the time value of money (paragraph 5.5.17).
Lifetime PD does more than size the provision. It also helps decide the stage. IFRS 9 asks banks to compare the risk of default over the expected life at the reporting date with the risk at initial recognition (paragraph 5.5.9). Many banks therefore track a relative change in lifetime PD against a threshold. Some use the 12-month PD as a proxy, but only when it is a reasonable approximation.
Days past due act as a backstop. IFRS 9 contains a rebuttable presumption that credit risk has increased significantly at 30 days past due (paragraph 5.5.11). As a result, a weak term structure can distort both staging and ECL at the same time.
Three related PD measures appear in every term structure. Confusing them is the most common source of error in lifetime PD work.
Lifetime PD is the cumulative PD at the end of the expected life, C(T). ECL calculations use marginal PD, because it ties to one specific period.
Assume a five-year loan with the annual conditional PDs below. All figures are illustrative.
| Year | Conditional PD h(t) | Survival S(t) | Cumulative PD C(t) | Marginal PD m(t) |
| 1 | 2.00% | 98.00% | 2.00% | 2.00% |
| 2 | 2.50% | 95.55% | 4.45% | 2.45% |
| 3 | 3.00% | 92.68% | 7.32% | 2.87% |
| 4 | 3.00% | 89.90% | 10.10% | 2.78% |
| 5 | 2.50% | 87.66% | 12.34% | 2.25% |
Percentages are rounded. Before rounding, the five marginal PDs add up exactly to the cumulative PD of 12.34%.
Here, the 12-month PD is 2.00% and the lifetime PD is 12.34%. In other words, lifetime PD is about six times the 12-month PD. Notably, you cannot simply multiply a 12-month PD by the number of years. Even a flat 2% annual hazard gives a five-year PD of 9.61%, not 10%, because survival compounds.
Banks typically choose among three approaches. The right choice depends on data depth, portfolio type and model governance.
| Approach | How it works | Data needed | Strengths and limits |
| Survival or hazard model | Estimates the hazard rate by period, often with a discrete-time logistic model using time on book and macro covariates | Loan-level performance history with default flags | Handles unmatured loans and macro covariates well. Needs long, clean history and is harder to explain |
| Markov transition matrix | Estimates migration between rating or delinquency states, with default as an absorbing state | Rating or bucket history across periods | Intuitive and consistent with staging. Assumes memoryless migration and has sparse cells at long horizons |
| Vintage or cohort curves | Tracks cumulative default rates by origination cohort and months on book, then extends to maturity | Cohort-level default history | Simple and transparent for retail loans. Responds weakly to macro shifts and relies on extrapolation |
These models estimate the hazard rate for each period. A discrete-time logistic regression is common, with time on book and macroeconomic variables as covariates. Because the model uses loans that have not yet matured, it makes good use of incomplete histories. However, it needs loan-level data.
Here, you estimate the probability of moving between rating or delinquency states. Default is an absorbing state, so a borrower cannot leave it. Let P be the one-period transition matrix. If P stays constant, the t-period cumulative PD for a starting state is the default-column entry of P^t. With forward-looking scenarios, P changes by period, so you multiply the period-specific matrices instead. This approach fits rated corporate portfolios well, and it links naturally to staging.
This method tracks cumulative default rates by origination cohort and months on book. Next, you extend the curves to maturity using a seasoning pattern. It is simple and transparent for retail loans. However, it responds poorly to macro shifts unless you add a regression layer or an overlay.
For retail portfolios with rich history, start with survival or vintage methods. For rated wholesale portfolios, transition matrices usually fit better. In every case, calibrate the first-year PD to your 12-month PD model, so that Stage 1 and Stage 2 numbers reconcile. Finally, backtest predicted cumulative PDs against observed cohort defaults.
Most teams already have a 12-month PD model. The steps below extend it to a full term structure.
Finally, backtest and document every assumption. Auditors and validators will ask for both.
IFRS 9 requires reasonable and supportable information about past events, current conditions and forecasts. Therefore, a term structure built only on long-run history is not enough.
A widely used technique is the Vasicek single-factor adjustment. For each period, the point-in-time PD is:
PD_PIT(t) = N( [ N^(-1)(PD_TTC) − √ρ × Z(t) ] ÷ √(1 − ρ) )
Here, N is the standard normal distribution function, N^(-1) is its inverse, ρ is the asset correlation, and Z(t) is a credit-cycle index. A negative Z represents a downturn and raises PD. Analysts usually link Z to variables such as GDP growth, unemployment and interest rates through regression. Alternatively, you can regress default rates on macro variables directly.
Many banks use an explicit forecast period of a few years and then revert to long-run averages. This is common practice, not a requirement, so document your choice.
IFRS 9 requires an unbiased, probability-weighted outcome. Banks therefore run several scenarios, commonly base, upside and downside. For example, suppose the 12-month PD is 2.0% in the base case, 1.5% in the upside and 3.5% in the downside. With weights of 50%, 20% and 30%, the weighted PD is 2.35%.
However, weighting PDs is only an approximation. ECL is non-linear in PD, LGD and EAD. As a result, many banks calculate ECL in each scenario first and then weight the results.
Models can lag fast-moving shocks. Consequently, banks add post-model adjustments, known as overlays, for risks the model misses. Document the rationale, approve each overlay through governance, and review it every reporting cycle.
Now connect the term structure to provisions. ECL is the sum, across periods, of marginal PD × LGD × EAD × discount factor:
ECL = Σ [ m(t) × LGD(t) × EAD(t) × DF(t) ] for t = 1 to T, where DF(t) = 1 ÷ (1 + EIR)^t
The assumptions below are illustrative.
| Year | Marginal PD | EAD (₹) | LGD | Discount factor | ECL (₹) |
| 1 | 2.0000% | 10,00,000 | 45% | 0.8929 | 8,036 |
| 2 | 2.4500% | 8,00,000 | 45% | 0.7972 | 7,031 |
| 3 | 2.8665% | 6,00,000 | 45% | 0.7118 | 5,509 |
| 4 | 2.7805% | 4,00,000 | 45% | 0.6355 | 3,181 |
| 5 | 2.2476% | 2,00,000 | 45% | 0.5674 | 1,148 |
| Total | 12.34% | 24,904 |
The total uses unrounded values, so it differs by ₹1 from the sum of the rounded rows.
A Stage 1 loan carries only the year-one amount, about ₹8,036. In Stage 2, lifetime ECL is about ₹24,904, roughly 3.1 times higher. This step-up is often called the cliff effect. It is smaller than the six-fold PD ratio because the balance amortises and later losses are discounted.
For term loans, the expected life follows contractual maturity, adjusted for expected prepayments. Revolving facilities such as credit cards differ. Paragraph 5.5.20 gives a limited exception for instruments that combine a loan and an undrawn commitment, where the bank’s contractual right to cancel does not limit its exposure in practice. In that case, the period reflects how long the bank is actually exposed to credit risk (see also paragraph B5.5.40). It can therefore exceed the contractual term. Banks usually estimate this behavioural period from historical experience and their credit risk management actions.
The example holds LGD flat to keep the arithmetic clear. In practice, LGD changes with collateral values, time to recovery and the macro scenario. Likewise, EAD changes with amortisation, prepayment and, for revolving lines, drawdown behaviour. Therefore, a lifetime PD is only one input to a period-by-period ECL.
Several errors recur across banks and NBFCs. Check your own models against this list.
Lifetime PD turns a single default estimate into a view across the whole life of a loan. First, build a term structure with a method that suits your data. Next, condition it on forward-looking scenarios. Finally, combine it with LGD, EAD and EIR discounting to produce ECL.
To see the next step in code, read our guide on how to build an ECL model in Python. If you prefer instructor-led learning, the PD, LGD and EAD in IFRS 9 section of our Advanced Certificate in IFRS 9 Modeling covers term structures, macroeconomic scenarios and scenario weighting. To ask about the next batch, email hello@dexlabanalytics.com.
Lifetime PD is the probability that a borrower defaults at any point over the expected life of an exposure. It is the cumulative PD at the end of that life. IFRS 9 uses it to measure lifetime ECL for exposures in Stage 2, where credit risk has increased significantly since initial recognition.
A 12-month PD covers defaults in the next year only, and it drives Stage 1 ECL. Lifetime PD covers defaults across the whole expected life, so it is never lower than the 12-month PD. For exposures with several years remaining, it is usually much higher.
A PD term structure is the set of default probabilities for each future period of an exposure’s life. Analysts show it as conditional, marginal or cumulative PD by year or month. They build it with survival models, transition matrices or vintage curves, then adjust it for forecast macroeconomic conditions.
Subtract the previous period’s cumulative PD from the current one. Marginal PD in year t equals C(t) minus C(t−1). For example, cumulative PDs of 2.00% in year one and 4.45% in year two give a year-two marginal PD of 2.45%. ECL calculations use this marginal figure.
Generally, yes. IFRS 9 asks for an unbiased estimate that reflects current conditions and reasonable, supportable forecasts. That points to point-in-time PD. Many banks therefore convert through-the-cycle PDs, built for Basel, with a credit-cycle adjustment such as the Vasicek single-factor approach.
Banks link PD to variables such as GDP growth, unemployment and interest rates. Next, they run base, upside and downside forecasts, each with a probability weight. Many then compute ECL in each scenario and weight the results. This gives the unbiased, probability-weighted estimate that IFRS 9 requires.
Understanding the difference between an application vs behavioural scorecard is the first step. Building one, from binning and WOE to validation and cut-off strategy, is where the skill really develops. If you want to learn this end to end, with real-world case studies and tools used in banks, explore our IFRS9 Certification course.
Explore Dexlab Analytics’ Credit Risk Modeling certification program to build PD, LGD, and EAD models from scratch, work through IFRS 9 ECL frameworks, and learn model validation techniques used by practicing risk teams.
.
Credit and Market Risk, Credit Risk, credit risk analysis, Credit Risk Analytics And Modeling, credit risk analytics training, Credit risk certification India, Credit risk modeling course India, Credit Risk Modelling, Credit Risk Modelling Using SAS, Credit scorecard modeling, Expected credit loss ECL, IFRS 9 modeling course India, Logistic regression scorecard, PD LGD EAD modeling, Python credit risk modeling
Comments are closed here.