Statistics

Research in Economics and Finance

Overview

At Kent there is particular interest in the use of nonparametric methods including quantile regression and Bayesian nonparametric approaches. Application areas include modelling of business cycle and capacity utilization, calculating sovereign credit ratings, modelling of stock return data, and predicting inflation.

Areas of research

Stochastic volatility models

Stochastic volatility models are a popular way of modelling financial time series. These models concentrate on capturing changes in the spread of observations over time.

Work at Kent has focussed on the use of nonparametric methods with these models to allow flexible specification of the dynamics and the distribution of the observations. These methods avoid making strong parametric assumptions about the data which can have important consequences for the ability of the model to predict well.

Current work is looking at adapting these models for high-frequency and ultra high-frequency (trade-by-trade) data.

Quantile regression

Quantile regression has a long history in econometric applications, where it can be used to address possibly evolving quantile-specific covariate. For example, the graph on the right is based on a study by Blair (1991) which examined the effect of experience on the annual salaries of statistical professors. Besides nonlinearity and some heteroscedasticity, what is also evident is the big difference in the effects of experience on the three quartiles. View the graph at full size.

Quantile regression also adapts well to the analysis of censored data, common in economic surveys, and has implications in dimension reduction.

Stochastic frontier analysis

Stochastic frontier analysis is a popular method for measuring the efficiency of firms and has been successfully applied to many organisations including hospitals, banks, and farms.

A fully efficient frontier is defined which, in the case of production, would represent the maximum level of production possible with a given set of inputs and a given level of technology. The efficiency of all firms is then measured relative to this frontier. Stochastic frontier analysis assumes that the efficiencies follow a distribution. There has been much debate about the choice of this distribution in the literature.

The group at Kent have looked at using Bayesian nonparametric methods to avoid choosing a family of distributions and extensions to cases where this distribution depends on firm-specific characteristics (such as management structure).

Time-varying variable selection

Regression models are often used in economics and finance to understand the relationship between variables. For example, it is often interesting to predict inflation from economic indicators (such as unemployment, inflation, or growth) or to predict equity premiums in financial markets.

Good predictions are often difficult because the relationships between these variables are often changing over time (perhaps due to the business cycle or changes to legislation).

The group has looked at Bayesian methods which allow both the effects of variables and the variables included in the regression to change over time.

Researchers working in this theme

Name Keywords
Prof Jim Griffin Stochastic volatility, Time-dependent variable selection, nonparametric inference
Dr Fabrizio Leisen Bayesian nonparametric inference, stochastic volatility
Dr Cristiano Villa Operational risk assessment, and copulas applied to risk evaluation
Dr Xue Wang Copulas

Estimated volatility of the FTSE 100 index from 4/1/2000 to 27/11/2013
The figure above shows estimated volatility of the FTSE 100 index from 4/1/2000 to 27/11/2013. Click on the image to view a full-sized version.

The figure below shows the effect of experience on the annual salaries of statistical professors. Click on the image to view a full-sized version.

The effect of experience on the annual salaries of statistical professors

School of Mathematics, Statistics and Actuarial Science (SMSAS), Sibson Building, Parkwood Road, Canterbury, CT2 7FS

Contact us

Last Updated: 20/05/2016