Let's be honest, constructing an economic database is not nearly as sexy as using the database to answer timely policy questions. It certainly won’t lead to phone calls from the press seeking interviews, and it will not help you get your name on the front of newspapers and magazine. You’ll even be lucky to get more than a blank stare from your spouse. Only your mother may feign interest before she changes the subject to when you might be having children. But, what I found in my dissertation is that it is an important, if overlooked, topic, and a nicely constructed database can help make sure the analysts appearing on magazine covers are there because of the brilliant work they did and not how utterly wrong they were. The database construction component of my dissertation resulted in a series of three journal articles that help answer the question: how shall I construct my database?
There is a long history of methods to construct a disaggregated database that balances 1) aggregate economic data with 2) detailed technology-level data. These “matrix balancing” methods generally consist of an objective function (e.g. minimize disparity between the two data) and constraints (e.g. maintain economic consistency). The most common methods include RAS, column cross entropy, and ad hoc methods. Ad hoc methods allocate values in some “informed” way rather than by optimization. In fact, matrix balancing is so well studied that often the precise methods used in constructing the database is only partial documented or not documented at all. The documentation only tends to focus on the data used and the innovative way of modeling using the disaggregate database. Very little attention is placed on how these matrix balancing methods affect the actual results from the model – until now.
My first paper, published in Energy Economics, used the same input data (GTAP and detailed technology data), but different matrix balancing methods to study the impact of the method on modeling results. We find that the mathematical properties of the method influences the results in predictable ways. We conclude that Minimum Sum of Column Cross Entropy and pro rata methods introduce bias in either the “row share” or “cost structure” in the database, which are important for some types of policy shocks. Therefore, these methods are ill-suited for the creation of a flexible database where future research questions are not currently known. We hope that this result leads to greater documentation of the methodology for database construction.
The second paper, published in Economic Systems Research, creates a novel matrix balancing method which allows for one of the more restrictive constraints of a common method to be relaxed. The RAS or cross-entropy method requires two constraints that serve to preserve row share and cost structure. When one of those constraints is removed, the preservation of these relationships falls apart. However, in the electricity problem we don’t have reliable data for populating one of these constraints. Forcing the constraint would introduce unreliable data into the database construction. Therefore, I redesigned the objective to specifically target row share and cost structure so that unnecessary and overly-restrictive constraint can be removed. We find that the new method, termed Share Preserving Cross Entropy (SPCE), preserves these important economic relationships better than RAS/cross-entopy and is identical to RAS/cross-entropy when the additional constraint is added back into the SPCE. This method was highly useful for the electricity sector disaggregation, and I hope that it finds its way into other disaggregation problems in the future.
Now that we answered “which methods are best?” and designed a suitable method for the specific problem of disaggregating the GTAP electricity sector, we went ahead and did it. The result, GTAP-Power, is free for those who subscribe the GTAP database and is the most electricity-detailed publicly-available CGE database. In the spirit of the findings from the first paper, the construction methodology, including matrix balancing method, is documented in the inaugural edition of the Journal for Global Economic Analysis, the successor to the GTAP Technical Paper Series.
The series of three papers push the envelope of database construction for I-O, SAM, CGE, and IAM modeling. The findings should be useful for many independent researchers, and the resulting database should be useful in opening new vistas in electricity-economic modeling for the GTAP community. The next step is to create a simple, pedagogical model so that the GTAP community can leverage the new database completely and launch their own in electricity-relevant research questions. I will be presenting GTAP-E-Power at the Annual Conference on Global Economic Analysis at the World Bank in June and hope to publish it to the GTAP community in the Journal of Global Economic Analysis.