The main aim of the present work is to study the effect of the actions of learning agents on the macroscopic variables. The framework considered is an economic one, in which there are individuals producing a single perishable good and facing the uncertainty that they don’t know the next selling price. Thus the firms have to form an expectation on the future price in order to produce efficiently. In this work I follow the well known Greenwald–Stiglitz model in which the bankruptcy risk induces a risk–averse behavior in the firms. I don’t find the aggregate dynamics simply taking the expectation of the mi- croscopic equations, as in the Representative Agent framework.On the contrary, due to the complexity of the system, I cannot find an analytical solution and I have to resort to simulation, as it happens for most agent–based models. In my model firms try to forecast the price inductively exploiting the avail- able information. This is a big difference respect to the Representative Agent framework in which agents, thanks to the perfect knowledge of the environment, process information deductively. I analyze two different scenario in which firms can choose between two fore- casting rules. The forecasted price affects the decisions of production since it enters into the production function along with the firm net worth. The switch- ing mechanism among the rules is represented through the usual multinomial logit, or Boltzmann weight. The dynamics of the aggregate production shows persistent fluctuations and it is strongly influenced by the learning process. Its growth rates have a kur- tosis of about 4–5, which is compatible with some finding about time series of the index of industrial production. Moreover, standard tests reject the null hypothesis of normality for them. Finally, an attempt is made to reproduce some features of the dynamics of the agents’ choices building a very simple model with the tool of the Master Equations. I write a Boltzmann–like equation which describes the evolution of the state(forecasting rules) distribution.
Lo scopo principale di questo lavoro `e studiare l’effetto sulle variabili macroe- conomiche aggregate delle decisioni prese da parte di imprese in grado di se- lezionare il comportamento, o la strategia, da adottare tra le diverse possibili. Il contesto economico `e quello del modello di Greenwald–Stiglitz in cui, a causa del rischio di bancarotta, le imprese producono una quantit`a proporzionale al loro patrimonio netto. A differenza del modello originale, in questa tesi la dinamica aggregata emerge naturalmente dall’insieme dei comportamenti individuali: piuttosto che essere ricavata dal valore atteso delle relazioni microeconomiche, la produzione aggregata `e data semplicemente dalla somma di quelle singole, senza assumere la presenza di un ‘agente rappresentativo’. Inoltre viene rilassata l’assunzione secondo la quale gli agenti sperimentano poca incertezza rispetto al prezzo di mercato futuro. Viene cos`ı descritto un meccanismo di selezione e di implementazione di regole di previsione per tale prezzo futuro. In base all’informazione che si rende mano a mano disponibile le imprese compiono una stima statistica del prezzo e dell’errore di previsione. In base alla stima dell’errore commesso gli agenti scelgono di volta in volta quella che, secondo la loro valutazione, risulta essere la regola vincente. Dalla simulazione numerica emerge una dinamica per la produzione aggre- gata caratterizzata da fluttuazioni persistenti e decisamente influenzata dal com- portamento degli individui. Vengono analizzati due diversi scenari, con due possibili regole di previsione in ogni scenario. Nell’ultima parte della tesi viene introdotto un semplice modello analitico con l’obiettivo di catturare alcune caratteristiche della dinamica comportamen- tale delle imprese.La tecnica matematica adottata `e quella delle master equations , equazioni che descrivono l’evoluzione temporale della probabilit`a di un sistema, composto da numerose unit`a, di trovarsi in un particolare stato. Questo approccio sembra essere promettente in quanto consente di descrivere in un quadro unico la dinamica e le fluttuazioni delle variabili macroeconomiche. Inoltre permette un’analisi del comportamento del sistema anche al di fuori dell’equilibrio.
Effetto del Learning sulle variabili macroeconomiche / Pompili, Valentino. - (2015 Feb 20).
Effetto del Learning sulle variabili macroeconomiche
POMPILI, VALENTINO
2015-02-20
Abstract
The main aim of the present work is to study the effect of the actions of learning agents on the macroscopic variables. The framework considered is an economic one, in which there are individuals producing a single perishable good and facing the uncertainty that they don’t know the next selling price. Thus the firms have to form an expectation on the future price in order to produce efficiently. In this work I follow the well known Greenwald–Stiglitz model in which the bankruptcy risk induces a risk–averse behavior in the firms. I don’t find the aggregate dynamics simply taking the expectation of the mi- croscopic equations, as in the Representative Agent framework.On the contrary, due to the complexity of the system, I cannot find an analytical solution and I have to resort to simulation, as it happens for most agent–based models. In my model firms try to forecast the price inductively exploiting the avail- able information. This is a big difference respect to the Representative Agent framework in which agents, thanks to the perfect knowledge of the environment, process information deductively. I analyze two different scenario in which firms can choose between two fore- casting rules. The forecasted price affects the decisions of production since it enters into the production function along with the firm net worth. The switch- ing mechanism among the rules is represented through the usual multinomial logit, or Boltzmann weight. The dynamics of the aggregate production shows persistent fluctuations and it is strongly influenced by the learning process. Its growth rates have a kur- tosis of about 4–5, which is compatible with some finding about time series of the index of industrial production. Moreover, standard tests reject the null hypothesis of normality for them. Finally, an attempt is made to reproduce some features of the dynamics of the agents’ choices building a very simple model with the tool of the Master Equations. I write a Boltzmann–like equation which describes the evolution of the state(forecasting rules) distribution.File | Dimensione | Formato | |
---|---|---|---|
tesi_pompili.pdf
Solo gestori archivio
Tipologia:
Tesi di dottorato
Licenza d'uso:
Non specificato
Dimensione
542.03 kB
Formato
Adobe PDF
|
542.03 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.