© Scott McDonald
Estimating Social Accounting Matrices
This work is joint work with Sherman Robinson and builds on work he started with Golan and Judge in the early 1990s not supported by GTAP.
The computer codes are free to use and distributed under Creative Commons BY-SA-NC 4 (https://creativecommons.org/licenses/by-nc-sa/4.0). They are provided without warranty that they are fit for purpose. The programmes are not supported, but unsupported training materials are available.
We welcome and appreciate suggestions about how these codes can be improved, especially if accompanied by code.
The SAMEST model is based on information theory as applied to contexts in which the statistical problem is ill-posed since there are typically more data points to estimate than available data, i.e., there may be negative degrees of freedom. There are potentially an infinite number of solutions to a SAM estimation problem that satisfy the basic condition that expenditures by agents equal the incomes to agents, that will be manifest as an equality of row and column totals. The problem is therefore how to select from among these solutions the most likely or least unlikely solution.
The data used by the programme take the form of a matrix of prior point estimates for the transaction values in the SAM, together with a wide selection of control totals and sub-matrix aggregates specified by the user. All point estimates, control totals and aggregates can be defined with or without error. The objective function in the SAMEST is cross-entropy.
The code is accompanied by a Technical document, a User Guide and a Guide to Building a Prior SAM. The code and documents are under ongoing revision, thence they all drafts. The code and documents are believe to be approximately correct at the dates in the code and documents.