Essays in economics and machine learning

Geiecke, Friedrich Christian (2019) Essays in economics and machine learning. PhD thesis, London School of Economics and Political Science.

Text - Submitted Version Download (17MB)

Abstract

This thesis studies questions in innovation, optimal policy, and aggregate fluctuations, partially with the help of methods from machine learning and artificial intelligence. Chapter 1 is concerned with innovation in patents. With tools from natural language processing, we represent around 4.6 million patents as high dimensional numerical vectors and find a rich underlying structure. We measure economy wide and field specific trends and detect patents which anticipated such trends or widened existing ideas. These patents on average have higher citations and their firms tend to make higher profits. Chapter 2 discusses an application of reinforcement learning to outcomes from causal experiments in economics. We model individuals who lost their jobs and arrive sequentially at a policy maker’s office to register as unemployed. After paying a cost to provide job training to an individual, the policy maker observes a treatment effect estimate which we obtain from RCT data. Due to a limited budget, she cannot provide training to all individuals. We use reinforcement learning to solve for the policy function in this dynamic programming problem. Chapter 3 turns to the analysis of macroeconomic fluctuations. It introduces a mechanism through which perpetual cycles in aggregate output can result endogenously. Individuals share sentiments similarly to diseases in models of disease transmission. Consumption of optimistic consumers is biased upwards and consumption of pessimistic consumers downwards. In a behavioural New Keynesian model, recurring waves of optimism and pessimism lead to cyclical aggregate output as an inherent feature of this economy. Chapter 4 concludes with a brief empirical investigation of newspaper sentiments and how they fluctuates relative to aggregate economic variables. Here the focus is not on contagion, but on the measurement of aggregate business and economic sentiment since around 1850. Using the archive of the New York Times, I build a historical indicator, discuss its properties, and possible extensions.

Item Type:	Thesis (PhD)
Additional Information:	© 2019 Friedrich Christian Geiecke
Library of Congress subject classification:	H Social Sciences > HC Economic History and Conditions H Social Sciences > HD Industries. Land use. Labor T Technology > T Technology (General)
Sets:	Departments > Economics
Supervisor:	Den Haan, Wouter J.
URI:	http://etheses.lse.ac.uk/id/eprint/4173

Actions (login required)

Record administration - authorised staff only