This dissertation seeks to contribute to the growing literature on Agent-Based Macroeconomics along three lines of research organised in three distinct chapters: (i) an ABM application to a specific macroeconomic problem; (ii) a methodological contribution; (iii) a critical review about open challenges still to be faced by ABM modellers; Although the three chapters can be read independently from each other, I would argue that they show a certain degree of complementarity. Indeed, some insights and findings presented in chapter 2 and 3 support at least partially the modelling strategy implemented in chapter 1, whereas others can be used in the future to refine and improve the model presented in chapter 1. The organisation of the chapters may seem unusual, in particular the choice to place the literature review at the end of the dissertation. This reflects the trajectory of my research: admittedly, when I started my PhD journey I was not fully aware of many critical issues affecting ABMs, however as my research advanced unresolved challenges stood in my way, some of which I decided to directly address in this dissertation, this is the case of chapter 2, others I decided to critically discuss, which is the case of chapter 3. In chapter 1 I propose a macroeconomic model suitable for studying technological innovations and structural change, moreover I provide an application of such model which elucidates a plausible and empirically sound mechanism leading from automation to job polarization. The model proposed in this chapter extends and complements models already well established in the literature, in particular it introduces heterogenous consumption goods, the possibility to model a non vertically integrated multi-sectors economy, and a novel production system in which different types of labor must be combined in the production process. The model, in its general version, is intended to be flexible enough to accommodate for future extensions and therefore to address a variety of research questions dealing with technological change, structural change, and labor market outcomes in terms of: aggregate employment, labor flows across sectors and skill specific reactions to technological shocks. In chapter 1 I also provide an application of the model to study how automation can lead to job polarization. The modelling strategy has been designed to be as close as possible to available empirical evidence on robots, sectorial workers skill distribution, and consumer preferences over differentiated goods. This allows to provide empirical grounding to some interesting and sometimes counterintuitive results. In particular, the model helps to understand a possible mechanism leading from automation to job polarization. As we will see, a sector specific labor-saving and skill-biased technological shock, which per se should depress the employment share of low skilled workers and increase the employment share of high skilled workers, can actually set in motion a chain of causal events leading to structural change at first and eventually to job polarization. In the chapter such mechanism will be explained in details, moreover it will be justified in light of available empirical evidence and its robustness proved by means of extensive sensitivity analysis. The second chapter is a methodological contribution, which tries to understand how to model expectations in ABMs. To begin with, the chapter tries to clarify which type of rationality is best suited in the ABM framework, maintaining that: (i) rational expectations a lá Muth are neither applicable, nor needed in the ABM framework; (ii) what is sufficient to achieve in ABMs is collective rationality, which simply implies that the aggregate mean forecasting error is on average zero, i.e. the economy as a whole is not systematically mistaken in making predictions. Therefore, under the assumption that collective rationality is sufficient in the ABM frame-work, the chapter studies the performances of different expectations formation mechanism within two agent based models. Moreover, I introduce a learning algorithm which combined to "static" expectations allows to update the otherwise fixed parameters contained in the expectation rules. Thus, the goal is to study whether it is possible to obtain aggregate unbiased expectations in an ABM framework. Since typically a macro ABM does not have a closed form solution, I rely on extensive computer simulations in order to assess the performances of different expectation formation mechanisms in different contexts. I do so in two macro environments: (i) a very simple and stylised model in which agents try to forecast a stationary variable and (ii) a full fledged macro ABM in which agents try to forecast a trended variable. In case (i) I designed a simple model in which a central bank set the interest rate and households try to forecast the one-step ahed inflation rate. In this context I will assess the performances of different expectation formation mechanisms across policy regimes and policy shocks. In case (ii) I will employ the model put forward in Caiani et al. (2016) augmented by technological innovation as in Caiani et al. (2019). This is a full fledged macro ABM which I use as a laboratory to assess different expectation formation mechanisms applied to firms trying to forecast future sales. The expectation rules employed in the following exercise are: naive expectations, where the expected value of a variable equals its past realisation, adaptive expectations, trend following, and social learning in the form of a very simple genetic algorithm. Moreover, I also employ hybrid expectations in which adaptive expectations and trend following are combined with learning. The last chapter discusses some of the main challenges for ABMs, in particular it deals with how to bridge models to real data and how to address the Lucas critique in ABMs. Overall, it is intended as a review of ongoing research on these issues, however I also sug- gest some possible ways to deal with the specific problems surveyed. The chapter concludes with a preliminary meta-analysis trying to assess the state of the art of currently widespread modelling practises, with respect to the challenges laid down in the chapter.

Questa tesi cerca di contribuire alla crescente letteratura sulla Macroeconomia basata sugli agenti lungo tre linee di ricerca organizzate in tre capitoli distinti: (i) un'applicazione ABM a uno specifico problema macroeconomico; (ii) un contributo metodologico; (iii) una revisione critica sulle sfide aperte che devono ancora essere affrontate dai modellisti ABM;

Three Essays in Agent-Based Macroeconomics / Fierro, LUCA EDUARDO. - (2020 Mar 19).

Three Essays in Agent-Based Macroeconomics

FIERRO, LUCA EDUARDO
2020-03-19

Abstract

This dissertation seeks to contribute to the growing literature on Agent-Based Macroeconomics along three lines of research organised in three distinct chapters: (i) an ABM application to a specific macroeconomic problem; (ii) a methodological contribution; (iii) a critical review about open challenges still to be faced by ABM modellers; Although the three chapters can be read independently from each other, I would argue that they show a certain degree of complementarity. Indeed, some insights and findings presented in chapter 2 and 3 support at least partially the modelling strategy implemented in chapter 1, whereas others can be used in the future to refine and improve the model presented in chapter 1. The organisation of the chapters may seem unusual, in particular the choice to place the literature review at the end of the dissertation. This reflects the trajectory of my research: admittedly, when I started my PhD journey I was not fully aware of many critical issues affecting ABMs, however as my research advanced unresolved challenges stood in my way, some of which I decided to directly address in this dissertation, this is the case of chapter 2, others I decided to critically discuss, which is the case of chapter 3. In chapter 1 I propose a macroeconomic model suitable for studying technological innovations and structural change, moreover I provide an application of such model which elucidates a plausible and empirically sound mechanism leading from automation to job polarization. The model proposed in this chapter extends and complements models already well established in the literature, in particular it introduces heterogenous consumption goods, the possibility to model a non vertically integrated multi-sectors economy, and a novel production system in which different types of labor must be combined in the production process. The model, in its general version, is intended to be flexible enough to accommodate for future extensions and therefore to address a variety of research questions dealing with technological change, structural change, and labor market outcomes in terms of: aggregate employment, labor flows across sectors and skill specific reactions to technological shocks. In chapter 1 I also provide an application of the model to study how automation can lead to job polarization. The modelling strategy has been designed to be as close as possible to available empirical evidence on robots, sectorial workers skill distribution, and consumer preferences over differentiated goods. This allows to provide empirical grounding to some interesting and sometimes counterintuitive results. In particular, the model helps to understand a possible mechanism leading from automation to job polarization. As we will see, a sector specific labor-saving and skill-biased technological shock, which per se should depress the employment share of low skilled workers and increase the employment share of high skilled workers, can actually set in motion a chain of causal events leading to structural change at first and eventually to job polarization. In the chapter such mechanism will be explained in details, moreover it will be justified in light of available empirical evidence and its robustness proved by means of extensive sensitivity analysis. The second chapter is a methodological contribution, which tries to understand how to model expectations in ABMs. To begin with, the chapter tries to clarify which type of rationality is best suited in the ABM framework, maintaining that: (i) rational expectations a lá Muth are neither applicable, nor needed in the ABM framework; (ii) what is sufficient to achieve in ABMs is collective rationality, which simply implies that the aggregate mean forecasting error is on average zero, i.e. the economy as a whole is not systematically mistaken in making predictions. Therefore, under the assumption that collective rationality is sufficient in the ABM frame-work, the chapter studies the performances of different expectations formation mechanism within two agent based models. Moreover, I introduce a learning algorithm which combined to "static" expectations allows to update the otherwise fixed parameters contained in the expectation rules. Thus, the goal is to study whether it is possible to obtain aggregate unbiased expectations in an ABM framework. Since typically a macro ABM does not have a closed form solution, I rely on extensive computer simulations in order to assess the performances of different expectation formation mechanisms in different contexts. I do so in two macro environments: (i) a very simple and stylised model in which agents try to forecast a stationary variable and (ii) a full fledged macro ABM in which agents try to forecast a trended variable. In case (i) I designed a simple model in which a central bank set the interest rate and households try to forecast the one-step ahed inflation rate. In this context I will assess the performances of different expectation formation mechanisms across policy regimes and policy shocks. In case (ii) I will employ the model put forward in Caiani et al. (2016) augmented by technological innovation as in Caiani et al. (2019). This is a full fledged macro ABM which I use as a laboratory to assess different expectation formation mechanisms applied to firms trying to forecast future sales. The expectation rules employed in the following exercise are: naive expectations, where the expected value of a variable equals its past realisation, adaptive expectations, trend following, and social learning in the form of a very simple genetic algorithm. Moreover, I also employ hybrid expectations in which adaptive expectations and trend following are combined with learning. The last chapter discusses some of the main challenges for ABMs, in particular it deals with how to bridge models to real data and how to address the Lucas critique in ABMs. Overall, it is intended as a review of ongoing research on these issues, however I also sug- gest some possible ways to deal with the specific problems surveyed. The chapter concludes with a preliminary meta-analysis trying to assess the state of the art of currently widespread modelling practises, with respect to the challenges laid down in the chapter.
19-mar-2020
Questa tesi cerca di contribuire alla crescente letteratura sulla Macroeconomia basata sugli agenti lungo tre linee di ricerca organizzate in tre capitoli distinti: (i) un'applicazione ABM a uno specifico problema macroeconomico; (ii) un contributo metodologico; (iii) una revisione critica sulle sfide aperte che devono ancora essere affrontate dai modellisti ABM;
ABM; contribution; review
ABM; contributo; rivisitazione
File in questo prodotto:
File Dimensione Formato  
Tesi_Fierro.pdf

Open Access dal 01/10/2021

Descrizione: Tesi_Fierro
Tipologia: Tesi di dottorato
Licenza d'uso: Creative commons
Dimensione 4.32 MB
Formato Adobe PDF
4.32 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11566/278619
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact