Filters
Availability
Publication Year
Resource Type
Faculty
School/Institute
Publication Search Results
Sort
Results per page
1 - 8 of 8
-
(2022)ThesisResearch in computational statistics develops numerically efficient methods to estimate statistical models, with Monte Carlo algorithms a subset of such methods. This thesis develops novel Monte Carlo methods to solve three important problems in Bayesian statistics. For many complex models, it is prohibitively expensive to run simulation methods such as Markov chain Monte Carlo (MCMC) on the model directly when the likelihood function includes an intractable term or is computationally challenging in some other way. The first two topics investigate models having such likelihoods. The third topic proposes a novel model to solve a popular question in causal inference, which requires solving a computationally challenging problem. The first application is to symbolic data analysis, where classical data are summarised and represented as symbolic objects. The likelihood function of such aggregated-level data is often intractable as it usually includes a high dimensional integral with large exponents. Bayesian inference on symbolic data is carried out in the thesis by using a pseudo-marginal method, which replaces the likelihood function with its unbiased estimate. The second application is to doubly intractable models, where the likelihood includes an intractable normalising constant. The pseudo-marginal method is combined with the introduction of an auxiliary variable to obtain simulation consistent inference. The proposed algorithm offers a generic solution to a wider range of problems, where the existing methods are often impractical as the assumptions required for their application do not hold. The last application is to causal inference using Bayesian additive regression trees (BART), a non-parametric Bayesian regression technique. The likelihood function is complex as it is based on a sum of trees whose structures change dynamically with the MCMC iterates. An extension to BART is developed to estimate the heterogeneous treatment effect, aiming to overcome the regularisation-induced confounding issue which is often observed in the direct application of BART in causal inference.
-
(2022)ThesisEssay 1. Since the GFC, banking regulators are increasing regulations to lower the risk of banking crises occurring in the future. These have focused on liquidity, asset composition, and capital requirements. This paper focuses on the loan rate effects of raising bank capital requirements. Previous calculations of loan rate effects predominately use the WACC formula, which implicitly assumes perfect competition. Banking is an industry where firms have market power. This paper develops a tractable model of an imperfectly competitive banking market, where the implicit cost of capital is naturally included in the optimization problem of the bank. I find that the magnitude of loan rate changes depends on the market structure and provide estimates for a calibrated Australian market. %This provides some validation of studies that use a blended MM approach. Essay 2. Using a tractable two-loan type banking model I analyse recent changes in capital regulation in Australia and NZ. My modelling shows, IRB banks were significantly advantaged by Basel II in generating ROE, but that subsequent changes in capital settings have slowly moved towards competitive neutrality. In addition, I analyse the minimum average risk weight policy for IRB banks (2016) and find that IRB banks are motivated to hold a higher proportion of risky loans and undoing composition efficiency. Moving back to risk-sensitive risk weights is desirable, but using a scalar multiple or correlation adjustment, reinstates the large advantage IRB banks have in low-risk loans. Lastly, I analyse RBNZ's proposed large increase in capital requirements using a two-country banking model. Due to NZ's dependence on IRB Australian banks, I find the loan rate impact of these changes depends on APRA's approach, with the loan rate effect much smaller than documented by the RBNZ. Essay 3. In Australia, mortgage brokers (MB) are paid by banks an upfront commission and a trail commission, rewarding the length of time a borrower stays with a bank. Both Hayne (2019) and PC (2018) recommend banning the trail commission but differ regarding who should pay the upfront commission, banks or customers. This paper uses a 2-period IO model of a mortgage market to analyse different MB remuneration options. I find if aggregator/MB firms are efficient, borrowers will benefit from banks' paying an upfront commission, with aggregator/MB firms no worse off and bank profits lower. In contrast, customers paying an upfront commission can be better for borrowers but will be worse for both aggregator/MB and banks. If a proportion of borrowers balk at paying MBs the commission, borrower gains are diminished, aggregator/MB firms are worse off, and some bank profits recouped.
-
(2022)ThesisThis thesis consists of three self-contained essays in development economics. A key theme common to all essays is the examination of various issues pertaining to economic development in less-developed countries from a microeconomic perspective. In the thesis, I explore methods of increasing human capital and the impact of negative economic shocks on mental health as well as behavioural preferences, focusing on adults and children alike. This is done with a view to highlight policies that can be implemented to improve outcomes for those living in low-income settings where access to resources is scarce and public safety nets are lacking. In the first essay, I assess the impact of a classroom-based positive psychology intervention on improving academic performance by conducting a Randomised Control Trial (RCT) with 899 school-aged children (10-14) years across 29 schools in Pakistan. I find limited evidence that the RCT had an impact on improving well-being measures. I find some evidence of improvement in the sense of agency in the short term, with those in the treatment group scoring 0.34 SD higher than those in the control group. There is also some evidence of improvement in the Mathematics, English and Urdu scores in the medium-long term; being in the treatment group is associated with 0.48 SD higher Mathematics scores, 0.32 SD higher English scores and 0.43 SD higher Urdu scores, and these results are statistically significant. I also find evidence that parental investment preferences modify the impacts of the intervention. Overall, the results suggest that young children require a support structure to succeed in their academic endeavours and while classroom-based interventions can improve learning outcomes to a certain extent, they need to be accompanied with the right support at home. In the second co-authored essay, we combine data collected just prior to the unfolding of COVID-19 with follow-up data from July 2020 to document the adverse economic effects of the pandemic and resulting impact on parental and child mental well-being in rural and semi-urban Pakistan. We find that 22% of the households in our sample are affected by job loss, with monthly income down 39% on average. Our difference-in-difference results show that job loss is associated with a 0.88 standard deviation (SD) reduction in adult mental health score (K10), a 0.43 SD reduction in a ‘Hope’ index of children’s aspirations, agency and future pathways, and a 0.39 SD increase in children’s depression symptoms. In addition, we observe higher levels of parental stress and anger reported by children, as well as an increase in reported prevalence of domestic violence. Overall, we document that the pandemic has disproportionately and negatively affected the economic and mental well-being of the most vulnerable households in our sample. In the third co-authored essay, we investigate the possibility that females and males had a distinct path in the evolution of competitiveness and cooperation. We conducted an experiment to elicit preferences for in-group egalitarianism and individual competitiveness for a random sample of 751 individuals in Sierra Leone (aged 18-85) to contrast the behavioural consequences of victimisation during the 1991-2003 civil war across gender and parental roles. Our data shows that mothers and fathers display the highest level of cooperation, yet conflict exposure does not affect them. Egalitarianism increases after victimisation only among non-parents, with the effect stronger for males who are the least egalitarian to start with. Conflict exposure tames everyone’s competitive tendencies, but has the opposite effect for mothers, the least competitive in the absence of conflict. A sample of competitiveness among 191 parents from Colombia shows a similar effect. Our results imply that conflict, by closing gender and parental gaps in behaviour, selects for pressures to reduce within-group differences possibly to enhance internal cooperation. It primes individuals towards group and individual survival depending on both gender and parental role. Overall, the thesis provides novel insights into some of the many challenges faced by low-income populations, including education and a lack of public safety nets. It also explores the consequences of negative economic shocks on mental health and behavioural preferences.
-
(2022)ThesisThis dissertation consists of three self-contained essays on Chilean Labour Economics. The first essay analyses the early access to pension funds in Chile during the COVID-19 pandemic, describing the long-term impacts and their distribution across the population. Using Monte Carlo simulations on the Chilean Social Protection Survey linked with administrative data, this study quantifies the effects of a 10% early release of pension funds. Each withdrawn dollar brings losses of 1.59 dollars in future retirement savings, reducing monthly pension benefits by 7.26%. This policy raises income inadequacy and inequality in retirement, increasing government expenditure by 4.33% to counteract these effects for 65-year-old retirees. Given the resulting increased pressure on welfare systems, I propose four policies to mitigate these effects and address the current challenges of most defined contribution pension schemes. Increasing contributions combined with an intra-generational solidarity component shows the biggest impacts. Contribution enforcement, reducing tax evasion, and delaying retirement by at least one year via incentives have lower but significant effects. The second essay investigates how the mix of private-public pension benefits and their design drive people's retirement choices. Using data from the Chilean Social Protection Survey linked with administrative information, this paper develops a structural Random Utility Model à la Stock and Wise (1990), to analyse the impact of social benefits and labour market conditions under a 3-option model (non-retired, partially retired, and fully retired). Results from conditional and mixed logit methods show that most workers retire shortly after the minimum statutory retirement age, sacrificing exponential increases in pension benefits. This choice is not due to a greater preference for leisure, but rather to the option of boosting current income by supplementing pension benefits with transitory labour income. These findings are crucial in the context of public policies implemented exclusively to supplement outcomes at the end of the working life with no structural changes to the pension system. Unlike standard 2-option models that find limited effects of public benefits, we find that a 20% increase in government supplements would lead to a 3.4% rise in retirement, which is mainly due to a 4.41% rise in partial retirement. Similar results emerge when analysing changes to labour market conditions or other retirement policies such as eliminating the duty to contribute beyond the statutory age. The final essay examines the wage growth stagnation in light of two of the most popular methods to decompose TFP growth; the nonparametric approach Free Disposal Hull (FDH) and the parametric approach called Stochastic Frontier Analysis (SFA). Considering those approaches and using index number theory, we define various families of indexes to decompose productivity and wage growth. Under the FDH approach, and since the technology set is assumed to be a cone, our results show a limited capability of the production technology of incorporating changes induced by input price variations and increases in allocative efficiency. Thus, despite the input mix index cannot be interpreted as an input price index, it is a very good proxy and explains almost exclusively the changes in wage growth. In contrast, the SFA approach allows higher levels of factor reallocation and variable return to scale, so the productivity increase (decrease) cannot be only attributed to technology (inefficiency) growths as they are under FDH. Then, the changes in wage growth are not closely followed by the input mix index, and the input mix index is best interpreted as a measure of changes in the relative proportions of primary inputs used in production that is induced by a change in relative input prices. Those results are based on the Annual Chilean Survey of Manufacturers (ENIA) and consider aggregate information from the manufacturing sector.
-
(2022)ThesisThis thesis explores the effectiveness of some government policies regarding retirement incomes, namely the Australian old age pension and mandatory retirement saving through superannuation. Using an overlapping generations model for a small open economy, I investigate the effect of introducing an adjustable (non-linear) taper rate to a means-tested public age pension. Unlike a linear taper rate, a non-linear taper rate can be adjusted without affecting eligibility for the pension. Both a progressive taper rate (that increases with income) and a regressive taper rate are considered. A strongly regressive non-linear test with a low average taper rate is marginally preferable to the best-performing linear test. Means testing the pension tends to favour higher income earners but introducing a non-linear test can reduce the welfare cost to lower and middle-income earners. The thesis then considers pension policies in the context of a population with Gul-Pesendorfer style self-control preferences, in which households experience temptation to over-consume in the near term and must exercise costly self-control to save for retirement. First, I consider the role of means testing a public pension and find that the benefits of the means test are smaller when self-control preferences are stronger as the expected cost savings are smaller than they would otherwise be, and the population increasingly values the pension as a commitment device. I also find that populations with higher self-control costs prefer lower taper rates on the pension means test, and a universal pension may be preferred if self-control costs are sufficiently high. Second, two- and three-period life-cycle models are used to examine the effectiveness of a policy maker directly mandating that households save for retirement. I find that increasing the contribution rate on a savings mandate always leads to a utility gain for households with self-control preferences provided voluntary household savings are not crowded out. For households without self-control preferences there are no utility losses provided voluntary savings are not crowded out. A savings mandate will not necessarily increase household savings unless voluntary savings are fully crowded out. The optimal choice of contribution rate was found to largely depend upon the agent’s discount factor rather than the intensity of the self-control preferences.
-
(2022)ThesisIn this thesis, I study the question of how a policymaker should make a decision when the information is dispersed among social agents whose characteristics are interdependent, i.e., one agent's type depends on both her own signal and others' information. There are two challenges for the policymaker to make the institution more flexible. First, the interdependence of the agents presents difficulty to characterize the optimal mechanism by its implied multidimensionality. Second, the policymaker may face uncertainty over the beliefs of the agents and the equilibrium they would choose. This thesis contributes to the design of more flexible institutions by developing novel tools and methodologies to tackle these challenges. This thesis also demonstrates several important applications of the new techniques. Chapter 1 introduces a novel multidimensional majorization concept and shows the ordinal, topological features and convexity of the relevant function space. The new concept and properties are used to extend Myerson's ironing method for multivariate ironing of interdependent preferences. In this chapter, the multivariate ironing technique is used to solve a class of quasi-linear adverse-selection models with one principal and multiple agents. As demonstrated in Chapter 1, the optimal solution involves a particular partition of the type profile space. In Chapter 2, however, we consider all allowable partitions and introduce an innovative approach to geometrise these. Allowable partitions, with refinement-coarsening partial order, form a poset that can be represented by a Hasse diagram. The poset can be transformed into its secondary polytope in Euclidean space by the characteristic function. Chapter 2 derives not only the optimal mechanism but also determines the payoff environment in which one particular partition is optimal. Chapter 3 further introduces an admission-control scheme and derives the optimal public-control mechanism. The chapter examines two applications where public control plays an important role – supply chain regulation and club goods provision – to find the optimal policies for maximising social welfare. The chapter also provides some numerical examples to illustrate the characteristics of the optimal mechanisms with and without admission control.
-
(2022)ThesisThis thesis explores the development and novel application of linear panel data methods that use latent grouping variables in the modelling of time-varying unobservable heterogeneity. The methods are tailored for use in microeconomic applications with observational panel data by: a) controlling for individual-specific intercepts; b) focusing on the economic interpretability of the time-varying heterogeneity component of the models; c) addressing the problem of estimating unknown group memberships across an unknown number of latent groups. The most general model studied also allows for latent group structures in the partial effects of observed covariates, where groups in the covariate effects can be independent from groups in the unobservable heterogeneity. Classical and Bayesian statistical methodologies are considered, with the main methodological contributions being in the development of Bayesian approaches. For the kinds of applications studied, the Bayesian methods are shown to have more favourable properties, both in principle and in practice. Empirical applications to retirement decumulation and smoking policy in Australia demonstrate how the methods developed in this thesis may be used to learn about economically meaningful latent behavioural patterns across a range of applications.
-
(2022)ThesisThe thesis develops efficient Bayesian estimation methods for Evidence Accumulation Models (EAMs), an important class of Cognitive Science models. Over the last twenty years, these are the dominant models of decision-making that consider decision time and are used to address important theoretical and applied questions in psychology. Most modern applications of EAMs include a hierarchical random effects structure for individual differences. Traditional Bayesian methods, based on Markov Chain Monte Carlo (MCMC), can be very costly for estimating hierarchical EAMs. Variational Bayes (VB) methods are an alternative to Bayesian MCMC methods and are increasingly used for approximate Bayesian inference in a wide range of challenging statistical models. VB methods can produce results ten or 100 times faster than exact methods such as MCMC. However, unlike MCMC, variational methods are approximate. Despite their strengths, VB methods are not widely used in psychological research. The first contribution of the thesis is to develop VB methods for the two benchmark EAMs: the linear ballistic accumulator (LBA) model and the diffusion decision model (DDM). In addition, an exact Bayesian estimation based on MCMC is developed for the DDM including random effects. Empirical studies show that, with careful choices of the approximating distribution and optimisation algorithms, VB methods can produce relatively accurate estimates of the posterior first moments much faster than MCMC. The second important contribution is to propose a novel statistical model selection technique based on VB called Cross-validation with Variational Bayes (CVVB). The CVVB method uses VB to estimate the hierarchical EAMs. It reduces the computational time significantly, making it possible to analyse and compare many complex EAMs in a reasonable time. The third contribution of the thesis is to extend both exact and approximate Bayesian estimation to EAMs with covariates. The methods developed in this thesis are applied to simulated data, and to real data from three highly cited experiments.