How can the productive capabilities of each municipality be unleashed taking into consideration the resources available to them? A first pass at this ambitious question begins by understanding the set of outputs a municipality is capable of producing. We answer this by discovering relationships between agricultural inputs and outputs and ask a relatively simpler question: how similar are agricultural outputs in terms of the inputs they use? Answering this question is made difficult by the fact that most UPAs cultivate just one or two crops. This may be a rational response to economies of scale. Given a plot of land and inputs, it may be easier to cultivate one crop on the entire land than plant a number of them with each requiring a different care regimen3. It may be that the inputs available only allow for a few types of crops.
In this paper, we use the rural census data from Colombia to build an agricultural product space capturing the similarities between outputs. We test the predictive power of the product space and use this to answer the question above. In section 2, we discuss the various sources of data and how they are merged, cleaned, and transformed before processing. In section 3, we look at some high level features of the dataset and how inputs, outputs, and land use are related. In section 4, we explore the mechanics of diversification.
We construct similarity and density matrices and show that they do indeed predict what a municipality produces. Finally, in section 5 we use Machine Learning algorithms and the density matrices to predict municipalities that are best suited to produce a given output. Further, we identify "missing" municipalitiesoutput pairs i.e. municipalities that should be producing a given output at high yield but currently are not. Finally, we summarize our findings and suggest areas for further work.
In this report we will be making extensive use of concepts described in more detail in the companion report "How Industry-Related Capabilities Affect Export Possibilities," especially with respect to Machine Learning techniques.