Témata prací (Výběr práce)Témata prací (Výběr práce)(verze: 384)
Detail práce
   Přihlásit přes CAS
Simulation-based estimation methods in financial econometrics: Analysis of performance and comparison
Název práce v češtině: Simulační metody odhadu ve finanční ekonometrii: Analýza výkonnosti a srovnání
Název v anglickém jazyce: Simulation-based estimation methods in financial econometrics: Analysis of performance and comparison
Klíčová slova: Multiagentní model, Behaviorální finance, Simulovaná metoda momentů, Simulovaná metoda maximální věrohodnosti, Bayesovské odhady
Klíčová slova anglicky: Agent-based modelling, Behavioural finance, Simulated method of moments, Simulated maximum likelihood, Bayesian estimation
Akademický rok vypsání: 2022/2023
Typ práce: diplomová práce
Jazyk práce: angličtina
Ústav: Institut ekonomických studií (23-IES)
Vedoucí / školitel: PhDr. Jiří Kukačka, Ph.D.
Řešitel: skrytý - zadáno vedoucím/školitelem
Datum přihlášení: 30.06.2023
Datum zadání: 30.06.2023
Datum a čas obhajoby: 18.09.2024 09:00
Místo konání obhajoby: Opletalova, O109, AULA Michala Mejstříka č. 109
Datum odevzdání elektronické podoby:30.07.2024
Datum proběhlé obhajoby: 18.09.2024
Oponenti: PhDr. František Čech, Ph.D.
 
 
 
Seznam odborné literatury
Chen, S., Chang, C., & Du, Y. (2012). Agent-based economic models and econometrics. The Knowledge Engineering Review, 27(2), 187-219. doi:10.1017/S0269888912000136

Franke, R., Westerhoff, F. (2011). Estimation of a Structural Stochastic Volatility Model of Asset Pricing. Computational Economics. 38. 53-83. 10.1007/s10614-010-9238-7.

Kukacka, J., Sacht, S. (2023). Estimation of heuristic switching in behavioral macroeconomic models. Journal of Economic Dynamics and Control, 146. https://doi.org/10.1016/j.jedc.2022.104585

Lux, T. (2022). Approximate Bayesian inference for agent-based models in economics: a case study. Studies in Nonlinear Dynamics & Econometrics. 10.1515/snde-2021-0052.

Platt, D. (2020). A comparison of economic agent-based model calibration methods. Journal of Economic Dynamics and Control, 113. https://doi.org/10.1016/j.jedc.2020.103859

Zhang, J., Zhang, Q., Li, Y., Wang, Q., (2023). Sequential Bayesian inference for agent-based models with application to the Chinese business cycle. Economic Modelling, 126, https://doi.org/10.1016/j.econmod.2023.106381

Zila, E., Kukacka, J. (2023). Moment set selection for the SMM using simple machine learning. Journal of Economic Behavior & Organization, 212, 366-391. https://doi.org/10.1016/j.jebo.2023.05.040
Předběžná náplň práce v anglickém jazyce
Motivation

Simulation-based estimation methods have emerged as valuable alternatives to classical approaches when closed-form solutions are unavailable and analytical solutions are infeasible. This is particularly relevant in the field of agent-based modeling (ABM), where the criterion function is often intractable (Chen et al., 2012). Consequently, the development of estimation methods based on simulation has become crucial in this domain.

Both frequentist and Bayesian methods have emerged in the field of ABM. Two prominent frequentist simulation-based methods that have demonstrated success are the simulated method of moments (SMM) and the simulated maximum likelihood (SML). The SMM has been applied effectively by researchers such as Zila and Kukacka (2023), who also extended the method by a non-arbitrary selection of moments using machine learning and demonstrated the method on the classical financial-based models. On the other hand, the SML offers a compelling alternative to the SMM, although it can be more complex, it does not suffer from arbitrary moment selection. Kukacka and Sacht (2023) show its practical applications in the macroeconomic ABM in the New-Keynesian model.

The main goal of the Bayesian approach is to estimate the posterior distribution of parameters. This typically requires prior distribution and likelihood. The missing analytical solution in the ABM means that the posterior is mostly estimated by Markov chain Monte Carlo (MCMC) or the Sequential Monte Carlo (SMC), SMC seems to be more robust especially when dealing with multimodal likelihoods that often arise in the ABM (Zhang et al., 2023). An alternative could be Approximate Bayesian Computation (ABC), a likelihood-free method that utilizes the comparison of the summary statistics between simulated and empirical data (Lux, 2022). Hence, it could be seen as the Bayesian counterpart of SMM.

The motivation for this thesis stems from the need to address the literature gap in the comparison of the simulation-based methods and to provide a thorough investigation into their relative merits and drawbacks. This research aims to shed light on the performance of these simulation-based estimation methods in the financial ABM. The findings will not only contribute to a deeper understanding of the strengths and limitations of these methods but also provide valuable guidance for researchers and practitioners in choosing the most appropriate simulation-based estimation method for their specific problem.


Hypotheses

1. Hypothesis #1: The Bayesian method provides more accurate parameter estimates compared to the frequentist methods when applied to agent-based models.
2. Hypothesis #2: The frequentist methods method exhibits greater computational efficiency compared to the Bayesian methods in estimating parameters for agent-based models.

Methodology

The methodology of this thesis involves a comparative analysis of the simulation-based estimation methods. The focus will be on two frequentist methods: simulated method of moments (SMM) and simulated maximum likelihood (SML). Additionally, two Bayesian approaches will be investigated: the likelihood approach with Sequential Monte Carlo (SMC) and the likelihood-free approach Approximate Bayesian Computation (ABC).

The performance of methods will be assessed based on various criteria, such as the accuracy of parameter estimates, asymptotic behaviour, and computational efficiency. The models will vary from simpler autoregressive and random walk models to the more complex, classical financial ABM such as Franke & Westerhoff (2011). The main programming language will be Julia. As the computational complexity is very high for this thesis, I will very likely use the computer cluster or cloud services.


Expected Contribution

The expected contribution of this diploma thesis is to bridge a gap in the literature by providing a comprehensive comparison of the most prominent simulation-based estimation methods. Although these methods have been successfully employed in various fields, there is a lack of studies directly comparing their performance and characteristics in agent-based modelling. Platt (2020) compared three types of simulated methods based on distance with Bayesian estimation with MCMC, where the Bayesian approach seems to be superior. Lux (2022) also demonstrated that Bayesian ABC could be more efficient than the standard frequentist method. Zhang et al. (2023) showed better performance for SMC over MCMC in a multimodal agent-based environment. The main contribution of the thesis will be utilising both distance-based and maximum likelihood-based methods in their frequentist and Bayesian variants. The thesis will be able to compare distance methods with likelihood methods, but also distinguish between frequentist and Bayesian estimation, shedding light on different approaches in agent-based modelling. Furthermore, the thesis will attempt to not only compare methods based on accuracy but delve into their asymptotic behaviour and computing complexity as well. By conducting a thorough analysis and comparison of these methods, this thesis will offer valuable insights into their respective strengths and limitations, providing a solid foundation for future research in this area.

Outline

1. Introduction
2. Literature review
3. Methodology
4. Results
5. Discussion
6. Conclusion
 
Univerzita Karlova | Informační systém UK