Computational Methods and Models in Macroeconomics
New developments in computational power and methods have introduced a wide variety of new tools for economic analysis. This dissertation explores whether these new advances can help us understand and analyze macroeconomic phenomena.
Chapter 1 shows how recently developed clustering methods can be helpful in identifying consumer groups based on a history of purchasing behavior. While traditional methods of customer segmentation rely on observable characteristics of consumers or the products they buy, the methods I use in this chapter rely instead on identifying groups of consumers who buy a similar set of products. If consumers who buy similar products are likely to have similar preferences, then clustering groups of consumers who buy similar products can potentially uncover groups of consumers who share unobservable characteristics that drive their preference structures without explicitly specifying those preference structures. Using simulations of a discrete choice logit demand system, I show that a density peaks clustering method can effectively uncover consumers with different preferences, using a series of examples to show how different assumptions impact the effectiveness of the clustering algorithm.
In chapter 2, I apply the methods of chapter 1 to attempt to improve measurements of the cost of living. In particular, I show that methods for measuring cost of living that rely on aggregate CES representative agents will often overstate the gains from new product varieties when groups of consumers have different tastes for products. Since the purchase data of consumers shows that there is substantial heterogeneity in the sets of products consumers buy, estimating inflation using a representative agent approach could produce biased estimates. However, the methods from chapter 1 can help to mitigate the bias from heterogeneity by grouping similar consumers based on their purchase history. I apply the method to a large panel of consumers and show that clustering consumers reduces the welfare impact of new product entry. Estimation on clustered data finds lower elasticities of product substitution and implies inflation rates about half a percentage point higher than a representative agent approach.
Chapter 3 offers a different connection between computational advances and macroeconomics by studying how visual tools can help to understand macroeconomic dynamics. I develop a visual, interactive model that allows the user to adjust parameters and observe the dynamics of the economy in real time. The model is agent based and assumes that agents make decisions based on defined heuristics rather than maximizing behavior. The core of the model draws inspiration from two sources. First, it models firm pricing behavior based on customer market models where firms attract customers by lowering prices and increase prices to earn profit on a larger customer base. Second, it assumes that firms quantity choices are determined largely by the level of demand they currently face and make adjustments on the basis of Keynesian inventory adjustments. A benchmark calibration of the model shows that the economy experiences endogenous cycles in key economic variables. Simple policy experiments show that the model responds to typical Keynesian fiscal and monetary policies.