New methodology for analysis of time series data
Time series are observations over time of one variable, for instance unemployment, GDP or consumption. The analysis of time series data can involve a number of challenges. Aziz Ali has investigated this in his dissertation and presents methods that are less sensitive to measurement errors in time series data.
In order to be able to come to correct and meaningful conclusions when working with time series data, one must first make sure to use the right statistical method. First of all, one must figure out whether the time series data that is to be analysed is stationary or non-stationary, which can be a challenge with many pitfalls. Stationary time series fluctuate around a constant value or a trend, while non-stationary time series tend to wander about and do not return to a constant value or trend.
“Financial time series data is most often non-stationary, but it is still important to establish whether a certain time series data is stationary or non-stationary in order to be able to choose what methodology to use for your analysis, for instance prognoses or investigating long-term links between different economic variables”, Aziz Ali explains.
Most methods used today to test hypotheses on whether a time series is stationary or non-stationary are sensitive to contamination of data with measurement errors or the occurrence of volatility in financial time series data. To wrongly dismiss the hypothesis of a non-stationary time series can lead to the conclusions drawn from the statistical analysis being incorrect. A further limitation demonstrated by many tests today is that the tools are not sharp enough (lack power) when working with smaller quantities of time series data (say, 250 or less observations).
“The goal with my research is to present methods that are easy to understand and apply for the analyst who is to decide whether a certain time series is stationary or non-stationary, but who suspects that the data has been contaminated by measurement errors”, says Aziz Ali.
Aziz Ali has earlier worked at Umeå University and SLU (Sveriges Lantbruksuniversitet), and also for 17 years with pharmaceutical research and development.
“My main interest is the application of statistics. Even though I don’t have any plans at the moment on continuing my career as a researcher, I will have a lot of use of my research studies in my job since statistics involves a lot of theory and the methods are constantly developing”, Aziz Ali concludes.
The public defence of the doctoral thesis will take place on March 16 at 1 pm in room weber, Växjö.
Dissertation title: On the use of wavelets in unit root and cointegration tests
Contact
Aziz Ali
aziz.ali@lnu.se
Carina Sörgårn
Communications officer
carina.sorgarn@lnu.se
+46 470-70 85 52