What it means
How long until the machine fails, the borrower defaults, the customer leaves? Survival analysis is the branch of statistics built to answer, and its trick is counting the unfinished stories.
The core problem is censoring: most of your machines have not failed yet, and ordinary statistics does not know what to do with them, while survival analysis credits their partial information honestly. The NIST reliability handbook treats these methods as the foundation of lifetime analysis: the survival function estimates the probability of lasting past any given time.
The Kaplan-Meier curve is the field's workhorse: a step-down plot that shows the surviving fraction over time, dropping at each observed event and carrying the censored cases forward. Hazard is the complementary lens: the instantaneous risk of the event now, given survival so far, which can rise with age, fall, or follow the bathtub of early failures and late wear-out.
Regression enters through proportional hazards: Cox's model asks which factors, credit score, maintenance schedule, contract type, multiply the hazard, without specifying its shape over time. Finance uses it everywhere quietly: default timing, prepayment modelling, customer lifetime value, and insurance reserves all run on survival machinery.
For a non-finance reader, survival analysis is the science of the waiting room: it estimates not just how many will be called, but when, and it learns even from those still waiting. The bathtub curve is reliability's famous shape: early failures from defects, a flat middle of random accidents, and rising wear-out at the end, three hazards in one lifetime.
Left truncation is the subtler data trap: borrowers who defaulted before your data began are missing, so the visible book is biased toward survivors from the start. Competing risks complicate the exit question: a loan can end by default, prepayment, or maturity, and modelling each exit requires treating the others as censoring, carefully.
Machine learning absorbed the frame: gradient-boosted survival models now beat Cox on raw prediction, while the old model keeps the interpretability that regulators ask for.
In practice
Real-world examples.
Example
A lender drops the average-lifetime approach because it discards every loan still paying. Instead it estimates survival curves that keep those loans in the at-risk group. The reserve model then reflects when defaults happen, not only how many.
Example
The hazard spike between months eight and eighteen concentrates collections in year two. The collections team shifts its effort to that window. Reserves are front-loaded to match the timing.
Example
Cox hazard ratios expose one channel failing at twice the book's rate at equal credit scores. The channel review moves from pricing to underwriting. The lender tightens checks at that channel's point of sale.
Formula
Calculation
Survival function S(t) equals the probability of lasting beyond time t; Kaplan-Meier estimates it as the running product of (at-risk minus events) over at-risk at each event time; hazard h(t) is the event rate among survivors, and Cox's model multiplies a baseline hazard by exp(covariate effects).
Worked example. A fictional lender follows 10 new loans. One loan defaults in month 3. One loan is censored in month 5, meaning it is still paying when the data ends for it. A second loan defaults in month 6.
- At month 3, 10 loans are at risk and 1 defaults, so the factor is (10 - 1) / 10 = 0.90 and S(3) = 0.90.
- Between months 3 and 6, one loan defaulted and one was censored, leaving 10 - 1 - 1 = 8 at risk at month 6.
- At month 6, 1 of the 8 defaults, so the factor is (8 - 1) / 8 = 0.875.
- S(6) = 0.90 x 0.875 = 0.7875, or about 78.75% of loans estimated to survive past month 6.
The censored loan is not discarded. It counts in the at-risk group until month 5, which is how the method uses partial information honestly.Case study
Seen in the real world.
This case study is fictional and illustrative. A made-up auto lender wants to know not whether loans default but when, because the loss reserves and the pricing both depend on the timing. Its analyst's first act is to reject the obvious approach, averaging the lifetimes of defaulted loans, because it throws away every loan still paying. The Kaplan-Meier curves change the credit meeting's vocabulary: default hazard spikes between months eight and eighteen, then flattens, so the loss reserve should be front-loaded and the collection effort concentrated in the second year.
The Cox model adds the why: hazard ratios show used-car loans from one channel failing at twice the rate of the book, holding borrower score constant, which redirects the channel review from pricing to underwriting. The censored cases prove their worth in the annual model validation: loans still paying contribute their months of information, and the model's predicted survival curve tracks the realised one within the confidence band. The CFO's summary to the board is the method's sales pitch in one line: we stopped asking how many loans die and started asking when, and the reserves got honest in the same quarter. The lender's pricing sheet now carries a hazard curve behind every rate, invisible to the borrower and decisive for the margin.
Watch out
Common mistakes.
- Ignoring censoring; dropping unfinished cases biases every estimate downward, and the method exists precisely to keep them.
- Confusing hazard with probability; the hazard is the conditional rate now, not the cumulative chance, and the two move differently.
- Assuming proportional hazards hold; if factor effects change over time, the Cox model's single multiplier lies, and diagnostics come first.
Questions
People also ask.
What is survival analysis?
Statistical methods for time until an event, handling censored cases where the event has not yet occurred by using their partial information.
What is the Kaplan-Meier curve?
A stepwise estimate of the surviving fraction over time, dropping at each event and carrying censored observations forward.
Where is it used in finance?
Default and prepayment timing, customer churn and lifetime value, insurance reserving, and reliability analysis of assets.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%