Examples of work
Assistance with Data Mining
ESQUANT assisted a client who was interested in applying Data Mining techniques in order to improve customer retention in their electronic business. We introduced the client to the Rattle Data Mining software and assisted in applying logistic regression and random forest methods to model their clients’ behaviour.
We were asked to review two reports concerned with the Australian Energy Regulator's estimate of the cost of equity and the determination of the weighted average cost of capital. Our analysis, using the rugarch package in R, showed that the EGARCH model used in one of the reports was incorrectly specified and the results presented in that report could not be reproduced. Our results agreed with the findings presented in the second report.
Design of Sampling and Testing Program
In order to establish estimates of various sources of variability, we statistically analysed quality control testing data from two manufacturers. We then evaluated the relevant Australian Standard on which the testing was based, and suggested amendments to that Standard. We derived the relevant formulae so that appropriate sampling and testing programs could be designed. We also developed an inter-laboratory trial to demonstrate that the various testing laboratories are compliant to each other.
Yield Curves, Zero Coupon Bonds and Par Value Yields
ESQUANT, in association with the Statistical Consulting Service in the Department of Econometrics and Business Statistics at Monash University and the Competition Economists' Group (CEG), worked on modelling of data supplied by CEG. We fitted Nelson-Siegel yield curves to four classes of bonds and provided methods to improve the estimation of the non-linear Nelson-Siegel model. We supplied standard errors to the estimated parameters and to the estimates of the Debt Risk Premium at 7 and 10 years maturity. In addition, we provided a number of refinements by using robust methods in the analysis of daily data - non-linear mixed effect models and zero coupon Nelson-Siegel yield curves to estimate par yield curves. Finally, we compared our results with those obtained using the methods applied separately by the Australian Energy Regulator and the Economic Regulation Authority (WA). Specifically, we showed that the joint-weighted approach favoured by the ERA (WA) was downwardly biased and gave less precise estimates than those that employed yield curve fitting.
Paid Parental Leave Survey
This is an ongoing project conducted by a PhD student in the Department of Social Work at Monash University. The project involves an analysis of data from a large survey carried out using Survey Monkey. Using some relatively advanced R techniques, ESQUANT assisted the student with cleaning the data and transforming the Excel spreadsheet produced by Survey Monkey into a consistent form, suitable for analyses in SPSS.
Evaluation of Acceptance Sampling Schemes