PPML Gravity Model
The PPML gravity model estimates bilateral trade or other flows by Poisson pseudo-maximum likelihood with exporter-time and importer-time fixed effects, keeping zero flows and remaining consistent under heteroskedasticity.
PPML 중력모형은 수출국·연도와 수입국·연도 고정효과를 넣은 포아송 의사최우로 양자 간 흐름을 추정하며, 0 흐름을 버리지 않고 이분산에도 일관된 추정치를 냅니다.
Did a trade agreement increase trade flows once distance and country-year effects are controlled?
거리와 국가별 연도 효과를 통제하고도 무역협정이 교역량을 늘렸는지 어떻게 추정하나요?
The flow itself, not its logarithm, is modelled with an exponential mean and high-dimensional fixed effects, so zero flows stay in the sample and the estimates are not biased by heteroskedasticity. Coefficients on logged variables are elasticities, dummy effects are (exp(b) - 1) x 100%, and standard errors are clustered by country pair.
When to use PPML Gravity Model
- Analysing bilateral trade, migration, tourism or commuting flows with many zeros.
- Estimating effects of distance or policy dummies with multilateral resistance terms as fixed effects.
- Log-linear OLS would drop zeros or be biased under heteroskedasticity.
When to use something else
- The data are not dyadic flows over origin, destination and time; use ordinary Poisson or panel count models.
- Policy timing is staggered and you need event-study identification; use dedicated difference-in-differences estimators.
Assumptions and requirements
- The conditional mean of the flow is exponential in regressors and fixed effects.
- Observations are independent across clusters (country pairs by default).
- Fixed-effect groups contain at least one positive flow (others are dropped).
What the analysis reports
- Coefficients (elasticities) with cluster-robust SEs and percentage effects
- Number of observations, zero flows and clusters
- RESET test of the functional form
- Pseudo R-squared
How to report it
A PPML gravity model of [flow] (N = [value] pair-years, [value] zero flows; exporter-time and importer-time fixed effects) estimated a coefficient of [value] for [variable] (cluster-robust SE = [value], p = [value]).
Related methods
- Poisson Regression
- Panel Fixed Effects Regression
- Difference-in-Differences
Run PPML Gravity Model on Quorant
- Data you need
- a table (CSV or Excel) with an outcome variable, predictor (feature) columns, a unit (entity) ID, a grouping variable and a time variable
- Price
- 600 credits for one analysis up to 10,000 rows (complexity level 4); larger or wider data costs more. Optional methodologist review: 2400 credits.
- Checking
- Checked by property tests in every deployment
- Before you pay
- A free pre-analysis diagnosis checks your data and gives an exact quote; nothing is deducted until you approve it.