31 May 2010

Resource rent tax: why GE modelling?


In its editorial today The Australian Financial Review comments:

The only honourable course for the government now is to release all of its RSPT-related modelling so the public – aided by independent experts – can decide.

While it is about it, the Government might like to explain to us why it chose to use a General Equilibrium model (the KPMG Econtech model) rather than an input-output model like Treasury’s own PRISMOD model on which Treasury normally undertakes its tax and price effect modelling.

I am no econometrician but I do know a little about mathematics.  What I know about mathematical models like a GE model is that they are, in mathematicians’ terms, models of complex non-linear systems in which everything is connected to everything else, leading to feedback loops all over the place. Classic examples of such systems are ecological systems, weather systems and economic systems. They are chaotic in a strict mathematical sense, leading to the mathematics of these systems being known in the popular literature as “chaos theory”.

One of the defining characteristics of these complex non-linear systems is acute (and I mean acute) sensitivity to initial conditions, leading to the notion that the beating of a butterfly’s wings in the Amazon can trigger a tornado in Texas.  This phenomenon was discovered by meteorologist Edward Lorentz in 1961. He was running weather simulations and decided to check something that was occurring part of the way through a particular simulation.  He re-started the calculation by feeding the output from his simulation as data for the new run, and noted to his surprise that the simulation very quickly diverged from the simulation he was checking.  On considering what was happening he realised that, whereas his normal input data was to six significant figures, the model’s output data was only to three significant figures.  If you restart the model with a number like 0.493 instead of 0.493127, the whole simulation goes off course.

It should be noted that these models are completely deterministic – there are no random elements in them, and you can calculate a unique set of results from the data input.  The problem is that a tiny variation in the initial data can produce a quite different unique set of results.

GE models have their uses, which mainly revolve around understanding the processes going on in the system – a small change in A produces a big change in B but only a modest change in C – but they are almost useless as predictive tools because they are so sensitive to the data they are fed. That is why we can never hope to predict the weather more than a week (at best) in advance – we can feed more and more data into bigger and bigger super-computers, but it will never be enough.

To return to the specific subject of the KPMG Econtech modelling of the RSPT, acute sensitivity to “initial conditions” includes of course acute sensitivity to the assumptions which are fed into the model.  AFR journalist John Kehoe has some interesting things to say about that on page 6 of today’s Australian Financial Review:

 It is believed Treasury directed KPMG Econtech to assume the 40 per cent RSPT would not distort mining investment and the modelling projections were arrived at independent of key design features of the tax, including the rate, uplift factor, depreciation allowances and transition arrangements.

These design features are all assumed to be “perfect” so only pure resource rents are taxed by the RSPT.

Of course what those of us in the real world want to know is whether or not the tax will in fact distort mining investment. We want to see modelling which demonstrates that, not modelling which assumes it.

To turn to the PRISMOD model, this was the model which was developed by the young Ken Henry and his team on the instructions of Treasurer John Kerin at the end of the Hawke era, and which was used to such devastating effect in the Keating era by Treasurer John Dawkins to destroy John Hewson’s Fightback! package.

So I would like to see what that traditional Treasury tool, the PRISMOD model, would show us about this proposed tax, and I would like to know why the KPMG Econtech model was chosen in preference to it.

2 comments:

Cameron Murray said...

CGA models are not assensitive to intial conditions as you make out, nor particularly prone to butterfly effects. Typically the opposite is true and they converge rapidly to an new final condition.

They essentially take and IO model as a starting point and allow price changes and flow-on adjustments to occur. IO can tell you the initial shock, ceteris paribus, while the CGE model allows people to adjust to new price/quantity conditions.

A large IO shock does not always imply large real impact on the economy after adjustments occur.

Nicholas Gruen said...

Paul, the result in the KPMG model is, as you've documented, in the assumptions, not the model.