Hazard models for duration data in listing time-on-market
Learn how hazard models describe listing time-on-market, how censored observations and days on market fit in, and how to read survival and hazard results.
Contents
- How the hazard function works
- Cox proportional hazards in practice
- Censoring and the Kaplan-Meier view
- Questions about listing hazard models
- What is a hazard model for duration data?
- Why do censored observations matter in time-on-market analysis?
- When should I use Cox proportional hazards instead of a parametric survival model?
- How does days on market fit into a survival analysis?
A hazard model for duration data in listing time-on-market estimates the chance that a listing sells or leaves the market at a given moment, given that it has stayed listed until then. The model treats time on market as a time-to-event outcome rather than a simple average, so each listing contributes information for as long as it is observed. That structure lets you compare listings side by side while accounting for listings that have not yet sold.
How the hazard function works
The hazard function gives the immediate rate of an event at a point in time for listings still on the market. A rising hazard means a listing that has lingered becomes more likely to sell or withdraw soon, while a falling hazard describes the opposite pattern. The hazard differs from the survival curve, which reports the share still listed after a given period. Reading both together shows when turnover speeds up or slows across a set of listings.
Cox proportional hazards in practice
Cox proportional hazards is a regression approach that relates listing features to the hazard without fixing the shape of time itself. Coefficients describe how a feature shifts the hazard relative to a baseline, so a positive value raises the chance of an event at any moment. The method handles censored observations directly, which matters when a listing is still active at the end of your data. Checking the proportional hazards assumption is a necessary step before trusting the coefficients.
Censoring and the Kaplan-Meier view
Censored observations are listings whose final outcome you do not see, such as active listings or withdrawn ones without a recorded sale. Ignoring censored listings biases any average days on market figure, because they are excluded exactly when they last a long time. The Kaplan-Meier estimator builds a survival curve that keeps censored observations in the calculation until they leave observation. Parametric survival models go further by imposing a distribution on the hazard, which helps when you want smoothed estimates.
Questions about listing hazard models
What is a hazard model for duration data?
A hazard model for duration data estimates the chance of an event at each moment for units still at risk. In listing time-on-market work the event is a sale or a withdrawal, and the units are individual listings. The model output describes how that chance changes with time and with listing characteristics.
Why do censored observations matter in time-on-market analysis?
Censored observations matter because listings still active at the end of your data carry real information about survival. Dropping them would remove exactly the listings that stayed longest on the market, which distorts any comparison. Hazard models and the Kaplan-Meier estimator keep censored listings in the analysis until they leave observation.
When should I use Cox proportional hazards instead of a parametric survival model?
Cox proportional hazards suits cases where you want to relate features to the hazard without assuming a specific time shape. A parametric survival model suits cases where you need a defined distribution for the hazard, for example to smooth or extrapolate survival. The choice depends on whether the proportional hazards assumption holds and on how much structure your data supports.
How does days on market fit into a survival analysis?
Days on market is the duration variable that survival analysis models as time until an event. The hazard function describes how the chance of a sale or withdrawal changes as days on market increase. Modeling days on market this way avoids the bias of averaging only the listings that already sold.