This thesis aims to model some aspects of the economic environment as complex systems, and to analyze the dependencies of such systems using in particular Agent-Based Modeling (ABM) and Artificial Intelligence. In the first chapter, we aim to model the dynamics of the labor market with an ABM, paying particular attention to the effects that education produces in the competition between individuals looking for a job and on wage’s formation. We create two types of agents that interact with each other and with the surrounding environment: workers and firms. The main feature of the model is that the workers' skills are randomly assigned and obtaining an educational qualification takes place within the model endogenously. Even the technological level of the firms is randomly assigned, and these conditions lead to obtain very similar results at each run. We show that, by modifying the starting conditions, in any analyzed scenario the low-skilled workers are the ones penalized the most both in the employment rate and in the wage amount, and we stress how important the investment in human capital is. Robustness checks confirmed the reliability of the obtained results. In the second chapter, we propose a Convolutional Neural Network combined with a Bidirectional Long Short-Term Memory Network (CNN-BiLSTM) for the forecasting of macroeconomic variables, analyzing 18 time series about the economy of the United States of America. In the third chapter, we develop a Deep Convolutional Generative Adversarial Network (DCGAN) for stock price forecasting. Considering both single-step and multi-step forecasts, the results obtained by our proposed architectures are promising and improve upon the considered baseline econometric models, both in the economic and financial context. We suggest that Artificial Intelligence (and in particular, Deep Learning) should be investigated and incorporated more by the economists, as we believe it can deliver excellent results, especially in an era where big data availability is growing more and more.
Questa tesi mira a modellare alcuni aspetti dell'ambiente economico come sistemi complessi e ad analizzare le dipendenze di tali sistemi utilizzando in particolare la Modellazione ad Agenti (ABM) e l'Intelligenza Artificiale. Nel primo capitolo ci proponiamo di modellare le dinamiche del mercato del lavoro con un ABM, prestando particolare attenzione agli effetti che l'istruzione produce nella competizione tra individui in cerca di lavoro e sulla formazione del loro salario. Creiamo due tipi di agenti che interagiscono tra loro e con l'ambiente circostante: i lavoratori e le imprese. La caratteristica principale del modello è che le competenze dei lavoratori sono assegnate casualmente e l'ottenimento di un titolo di studio avviene all'interno del modello in modo endogeno. Anche il livello tecnologico delle imprese è assegnato casualmente, e queste condizioni portano ad ottenere risultati molto simili ad ogni run. Mostriamo che, modificando le condizioni di partenza, in qualsiasi scenario analizzato i lavoratori poco qualificati sono quelli maggiormente penalizzati sia nel tasso di occupazione che nel livello salariale, e sottolineiamo quanto sia importante l'investimento in capitale umano. Controlli di robustezza hanno confermato l'affidabilità dei risultati ottenuti. Nel secondo capitolo presentiamo una Convolutional Neural Network combinata con una Bidirectional Long Short-Term Memory Network (CNN-BiLSTM) per la previsione di variabili macroeconomiche, analizzando 18 serie storiche dell'economia degli Stati Uniti d'America. Nel terzo capitolo, sviluppiamo una Deep Convolutional Generative Adversarial Network (DCGAN) per la previsione del prezzo delle azioni. Considerando sia le previsioni single-step che multi-step, i risultati ottenuti dalle nostre architetture proposte sono promettenti e forniscono risultati migliori rispetto ai modelli econometrici di base considerati, sia nel contesto economico che finanziario. Suggeriamo che l'Intelligenza Artificiale (ed in particolare, il Deep Learning) dovrebbe essere studiata e incorporata maggiormente dagli economisti, poiché riteniamo che possa fornire risultati eccellenti, soprattutto in un'era in cui la disponibilità di dati sta crescendo sempre di più.
ESSAYS ON COMPLEX ECONOMIC SYSTEMS AND ARTIFICIAL INTELLIGENCE
Staffini, Alessio
2022
Abstract
This thesis aims to model some aspects of the economic environment as complex systems, and to analyze the dependencies of such systems using in particular Agent-Based Modeling (ABM) and Artificial Intelligence. In the first chapter, we aim to model the dynamics of the labor market with an ABM, paying particular attention to the effects that education produces in the competition between individuals looking for a job and on wage’s formation. We create two types of agents that interact with each other and with the surrounding environment: workers and firms. The main feature of the model is that the workers' skills are randomly assigned and obtaining an educational qualification takes place within the model endogenously. Even the technological level of the firms is randomly assigned, and these conditions lead to obtain very similar results at each run. We show that, by modifying the starting conditions, in any analyzed scenario the low-skilled workers are the ones penalized the most both in the employment rate and in the wage amount, and we stress how important the investment in human capital is. Robustness checks confirmed the reliability of the obtained results. In the second chapter, we propose a Convolutional Neural Network combined with a Bidirectional Long Short-Term Memory Network (CNN-BiLSTM) for the forecasting of macroeconomic variables, analyzing 18 time series about the economy of the United States of America. In the third chapter, we develop a Deep Convolutional Generative Adversarial Network (DCGAN) for stock price forecasting. Considering both single-step and multi-step forecasts, the results obtained by our proposed architectures are promising and improve upon the considered baseline econometric models, both in the economic and financial context. We suggest that Artificial Intelligence (and in particular, Deep Learning) should be investigated and incorporated more by the economists, as we believe it can deliver excellent results, especially in an era where big data availability is growing more and more.File | Dimensione | Formato | |
---|---|---|---|
Thesis PhD_Staffini.pdf
accesso solo da BNCF e BNCR
Dimensione
6.05 MB
Formato
Adobe PDF
|
6.05 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/159219
URN:NBN:IT:UNICATT-159219