Estimation of gravity models by OLS estimation, NLS estimation, Poisson, and Neural Network specifications

• Four specifications of gravity models for freight flow prediction are compared. The traditional specification with OLS estimation is compared with non-linear least squares (NLS) estimation as well as with a model where data are assumed to be Poisson distributed. These are compared with a semi-parametric neural network model. Data consists of freight flows between Norwegian counties. The attribute describing a node is population while distance gives the friction on links of transportation. Results show that estimation with OLS and NLS is inferior to Poisson and Neural Network specifications. However, the Poisson model, although advantageous compared to OLS, may still be improved upon. The semiparametric Neural Network does require less of these restrictions to hold and also outperforms the others as a tool for forecast in terms of Root Mean Square Error (RMSE). The NLS model although showed the best performance when estimated on known data.

Subject headings

SAMHÄLLSVETENSKAP  -- Ekonomi och näringsliv -- Nationalekonomi (hsv//swe)

SOCIAL SCIENCES  -- Economics and Business -- Economics (hsv//eng)

Keyword

Gravity model

Transportation

Freight flows

Spatial interaction

OLS

Poisson-regression

Non-linear regression

Neural network

Economics

Nationalekonomi

Publication and Content Type

vet (subject category)

rap (subject category)