Putting the pieces together
Let’s start our last course week by shortly reviewing how we started in the first week.
We explained that modeling can be structured into four main phases:
In the last weeks, we explored how to transfer your problem into different kinds of models (do you remember what the differences between Emission Choice Model (ECM), Output Abatement Choice Model (OACM), and Input Choice Model (ICM) are?). We addressed the challenges of ensuring that your model is solvable (do you remember the ‘as many equations as variables’ rule?). And we explored how to make economic sense out of those mathematical equations (do you remember how activity variables, costs, and prices were linked?). And you could explore different modeling elements with the little exercises along the way.
However, what we did not cover in detail in the last weeks is how to transfer all this knowledge into an actual computer model that can produce the numbers you have seen in the exercises. There is a large diversity in modeling software, coding approaches, and computer systems out there; however and a detailed lecture on numerical computational modeling is beyond the scope of this course. Nevertheless, we want to provide you with some basic guidelines on numerical modeling in the following steps: namely, data handling and computational restrictions.
Data handling has large impacts on the model design and the interpretation of the model and its results. While a large share of modeling work is done with pen and paper (especially conceptual assessments can often be carried out without a computer using purely illustrative data), quantifications, simulations, and scenario analyses will need a computer and numerical model implementation. And as you seldom will get all the data you would like to get, you will likely circle between your problem and the modeling phase until you have a model design that can achieve both: answer your research question and rely on available data. Similarly, you model results will drive what and how you can interpret into your model.
Computational restrictions have a large impact on the solution phase and thereby again feedback effects on your model design choice. As you will see in the next steps, even with modern computational capabilities, there are still some problem types that are hard to solve, especially for large-scale models. It is likely that the model design you would need to solve your initial research questions is simply not solvable. So, you will be forced to make adjustments, simplifications, or vary your assumptions to account for those constraints.
You see that the four modeling phases are indeed not as linear as the picture above suggests. Until you will have a running model with interpretable results you will likely have circled back and forth through multiple iterations of the four phases. That is perfectly normal. The more experience you get with modeling the easier it will become to move along those phases; even experienced modelers seldom get a perfect model up and running on their first attempt.
And finally, we encourage you to go out and start designing your own models. Building models, applying modeling software and writing model codes can only be learned by doing it.
© University of Basel