Welcome to NNE

This website provides the code and some documentation for the neural net estimator (NNE), a general approach that uses neural nets to estimate structural econometric models.

Please select the type of NNE below.

Original NNE (limited-information)

Based on “Estimating Parameters of Structural Models Using Neural Nets,” Wei and Jiang (2025), Marketing Science, 44(1). This NNE uses researcher-specified moments as input to the neural net. It is most useful when the researcher has clear intuition about what data moments identify the structural model.

Open →

Pre-trained NNE

Based on “Pre-Training Estimators for Structural Models: Application to Consumer Search,” Wei and Jiang (2025). This NNE pretrains a neural net for a given structural model, so researchers can use it to estimate the structural model right away — as easy as running a reduced-form regression.

Open →

Full-information NNE

Based on “Estimating and Assessing Identification of Structural Models via Deep Learning,” Wei and Jiang (2026). This NNE uses the whole dataset (instead of researcher-specified moments) as input. It can automatically exploit variation in data and thus is useful for not only estimating but also assessing the identification of a structural model.

Open →

	Original NNE	Pre-trained NNE	Full-information NNE
Input to the neural net	Researcher-specified moments \(\boldsymbol{m}\)	Regression coefficients & summary statistics	The whole dataset \(\mathcal{D}\)
You provide	Your structural model + moments	Just your data (search model)	Your structural model
Training needed	Yes (you train it)	No (pre-trained)	Yes (you train it)
Code	AR1 & Search	Search	Mixed Logit & Search