Senator JOYCE—I think the first thing we have to differentiate between is recommendation and options. If I were to give you an option on a number of cars, that is not as powerful a statement as if I were to recommend a certain car to you, is it? I refer obviously to the report where you give options, not recommendations.

Talk
Mr Samuel—I think you will find if you go to the conclusion there that we outlined some options. We dismissed two of the options, which were what I call the ‘Informed Sources option’ and then the modification of that which is to produce the Informed Sources information via or through the ACCC, and came down in favour of the Fuelwatch option, with some caution that some areas needed further examination which is what we have been addressing this morning.

Talk
Senator JOYCE—The endogenous breakpoint analysis modelling has some flaws, doesn’t it?

Talk
Dr King—None that I am aware of, but I am happy for you to enlighten us.

Talk
Senator JOYCE—Do you have any alternate statistical modelling that could have been provided on which to premise your decisions?

Talk
Dr King—Are you talking about the exogenous breakpoint analysis?

Talk
Senator JOYCE—Yes.

Talk
Dr King—Do you mean the exogenous breakpoint analysis that has been done here on weekly, monthly, average data and minimum data?

Talk
Senator JOYCE—Yes. Are there any other alternate modelling processes that could have been used instead of that form?

Talk
Dr King—Are there any other econometric tests that could have been used?

Talk
Senator JOYCE—Yes.

Talk
Dr King—There are a variety of econometric tests that can always be used in any situation. It is a matter of judgement by the modeller to choose the appropriate test and it is the commission’s view that this is the appropriate test to use on this data.

Talk
Senator JOYCE—Why is it the appropriate test?

Talk
Dr King—Perhaps it is best to start off by explaining what structural break analysis actually does so that we can understand why this is the appropriate test. When you are looking at any time series of data, such as we have here, a structural break looks at the underlying regime generating the data. Your data will have a profile, if I can put it that way, which may alter due to a specific underlying event. What structural break analysis does is it takes that data and checks against the possibility of there being no breaks or whether there are statistically significant breaks in that time series—in this case either one or two.

Talk
Senator JOYCE—Your premise of statistical significance is what?

Talk
Dr King—The standard test of statistical significance is based on the assumption of the normal distribution of errors and a 95 per cent or 99 per cent confidence interval.

Talk
Senator JOYCE—You are using a 95 per cent confidence interval?

Talk
Dr King—I can give you the PBS, if you prefer, which allows you to do your own confidence—

Talk
Senator JOYCE—Why do you use a 95 per cent confidence interval?

Talk
Dr King—That is the standard one used in econometrics, but as I said I am happy to give you the PBs.

Talk
Senator JOYCE—You did not in page 4 but you have just told me that you are going to actually do that now—is that correct?

Talk
Dr King—To give you the PBs?

Talk
Senator JOYCE—Yes.

Talk
Dr King—I am happy to, if you wish for me to run through them now.

Talk
Senator JOYCE—Obviously if the statistical modelling is questionable then the premise of the decision to take on Fuelwatch is questionable, isn’t it?

Talk
Dr King—I am sorry, this particular test is doing one very specific thing and obviously it is part of a number of tests that the commission undertook, some of which were reported in our December 2007 report and others are mentioned on page 3 of this press release. This particular test is simply analysing whether there are structural breaks in the data, determining whether those structural breaks are statistically significant and when those structural breaks occur. The aim of this test was to confirm that the FuelWatch effect was separate from, for example, a Coles entry effect into Western Australia, which is what this test does. This test shows that, yes, it was statistically separate from a Coles effect, so from this test we can say that FuelWatch has a statistically significant separate effect from Coles.

Talk
Senator JOYCE—Do the structural breaks relate to a better price?

Talk
Dr King—The structural breaks relate to the price series, yes.

Talk
Senator JOYCE—Do they relate to a better price?

Talk
Dr King—The coefficients are negative, so they relate to a price reduction, yes.

Talk
Senator JOYCE—With your analysis, though, you looked at posted prices, didn’t you?

Talk
Dr King—The data that we have is based on the posted prices, yes.

Talk
Senator JOYCE—It is looking at posted prices?

Talk
Dr King—Yes, it is.

Talk
Senator JOYCE—Why did it not look at actual prices?

Talk
Dr King—I am sorry?

Talk
Senator JOYCE—Is it appropriate from an economic testing point of view for the ACCC modelling to only look at the posted price and not do a full analysis on the price actually paid by motorists and the volumes sold at those prices?

Talk
Dr King—I think there are two parts to your question. The first is, of course, the posted prices are the prices paid by the motorists. Last time I went and filled my tank the posted price was the price I paid, and I suspect that they would be in some significant trouble if they started selling at anything other than the posted price. The price data used by the commission are the prices paid by the motorists. If the commission did not have available to it—and as far as I am aware the volume data for any period pre-2006 does not exist—

Talk
Senator JOYCE—The volume data does not exist?

Talk
Dr King—As far as I know.

Talk
Senator JOYCE—How does that affect the weighting of your modelling? If the volume data does not exist, does that start to diminish the veracity of your modelling?

Talk
Dr King—No, not at all. For modelling checks, the study used the relative prices between Perth before FuelWatch and after FuelWatch, using the east coast as a baseline. The results are here. I will not repeat them, but it shows statistically—

Talk
Senator JOYCE—At what point in time did volumes become a part of the modelling?

Talk
Dr King—Volumes are not in the model that we are looking at, the endogenous structural breaks. This is a posted price time series. You cannot simply change a time series half-way through, because you would end up with inconsistent data over time.