DSGE models can be even worse in practice

by
Editor
My articles My siteMy booksMy videos
Follow on:Twitter

from David Richardson

There have been numerous postings on this blog that go the problems associated with dynamic stochastic general equilibrium models. Most of these posts have questioned their use in the hands of the high priests of the orthodox economics profession. For example, Lars Syll has said “DSGE models are simply — from both empirical and methodological points of view — not suited for understanding modern monetary economies” (28 Nov 2018). I have no argument at all with this assessment but point out that what is done by the high priests is much ‘cleaner’ than the modelling specifically designed for partisan purposes.

There is a good example in Australia at the moment of the abuse of modelling for political purposes. Australia is in the midst of an election campaign which is mainly fought between the two major parties, the Labor Party and the governing Coalition of the Liberal Party and National Party. The two Coalition parties are shades of conservative and tend to include both climate change deniers and those that believe Australia need do nothing because it is too small to make a global difference.

A modeller who is sympathetic to the Coalition and has produced work for them in the past has entered the debate with his own modelling of the two parties’ climate policies and especially their emissions reductions targets (the Coalition target is a 26 per cent reduction on 2005 levels by 2030 compared with Labor’s target of 45 per cent reduction by 2030).

Brian Fisher (2019a and b) is the modeller in question and has produced two papers comparing the parties, one before the full details of Labor’s policy was announced and a second one purporting to model Labor’s specific policies. The results suggest Labor’s policy could result in a 6 per cent reduction in GDP, a massive reduction in wages and a carbon price of up to A$405 per tonne or about ten times the European price using the exchange rate as this is being written. These are now being used in the conservative press to scare Australian voters, especially in the Murdoch papers.

The modelling is never clear and limited detail is provided so we had to re-engineer the results to some extent. Both the Government and Opposition policies were a cap and trade type of system. The exact details do not matter, but the important thing is the government sets quantities and companies have to achieve those any way they can including by trading carbon credits. Fisher uses the phrase ‘shadow price’ so what Fisher seems to have done was ask what increase in carbon prices would give the quantities each party sought.  But Fisher forgets to mention that shadow prices are called just that because they are not actual prices. Indeed, a reduction in electricity usage should be associated with a fall in prices if anything. So the results completely miss the point of the policies; the intent of each policy is to set quantities but allow trading in carbon credits. This is completely different to setting a tax to change behaviour.

But Fisher’s ‘shadow’ prices are then fed back into the model to generate rather scary stories about loss of GDP, reductions in wages, and most other things we can think of. What Fisher has done is not give us the implications of rationing but of the alternative question (that no one asked): what tax on electricity/carbon would need to be imposed in order to generate the rationed quantities? To pretend shadow prices are actual prices is a major sleight of hand.

The model itself

Fisher’s model has at is core a production function, a constant elasticity of substitution (CES) production function in which electricity/energy is an important argument.  A production function itself is simply a mathematical statement that expresses the outputs of production as a function of the inputs used in production. A property of the CES production function is that the additional costs of substituting for the removal of one unit of input start out small at first but soon accelerate so that, for example, the cost of a one per cent reduction to the 26^th percentile is much less than a one per cent reduction to the 45^th percentile.  That seems to drive the results showing that a doubling of the target much more than doubles the ‘costs’.

The first major problem is that electricity is modelled as a ‘technology bundle’ with limited substitution between the different generation technologies; coal, hydro, wind and solar in particular. This conveniently stops all electricity being produced by one or two technologies (e.g. solar or wind) in the long run and prevents emissions reductions other than by reducing electricity production across the board. But of course the long run is precisely where this sort of modelling takes us and Fisher should have pointed out that any results he comes up with are temporary outcomes and on the way to some equilibrium with lower costs. As it is his assumptions have strange implications.

If demand dried up, as a result of the cap or otherwise, it is likely, given the relative costs, that fossil fuel generation would be shut down first but that cannot happen in the model. The limited degree of substitution is an ad hoc assumption that is chosen for convenience but certainly at the cost of realism. For example it implies that no matter what the relative prices, some coal generation will always be used and it is not completely substituted out of the technology mix.  Fisher’s treatment would also seem to imply that as customers cut back on electricity, the electricity producers would necessarily also cut back on cheaper renewables. This is a consequence of Fisher’s assumptions. But these implications seem bizarre in an exercise that attempts to model a clamping down on emissions. The one thing about electricity is that electrons are homogenous and neither producers nor users are constrained in their substitution between different sources.

Another major problem is the use of the CES production function which assumes that all businesses face a smorgasbord of possible ways of making their output; but with the property that a change in input prices will induce production to be less intensive in the factor price/s that increase/s. This is not necessarily true and we know that there are strict limits on how much electricity can be substituted for in the production of things like aluminium.

An important property of this model also seems to be its general equilibrium character. It is assumed that the economy is in equilibrium and such things as emissions targets disturb that equilibrium. But obviously Australia and the rest of the world are in a transition phase towards cheaper renewable energy sources.  Likewise, macroeconomic effects of induced investment in renewables are ignored and only things like ‘integration costs’ are modelled. These are the costs resulting when ‘increased investment in intermittent renewable electricity technologies incurs additional capital integration costs to firm generation from these sources’. But modelling should not ignore induced investment as a boost to aggregate demand or at the very least should justify any omission.

These are some of the main problems in the Fisher model and we could continue for example with the general problems of general equilibrium modelling. But hopefully enough has been discussed to show Fisher’s model conveniently skew his results towards vastly exaggerating any costs of Labor’s policies.

Even when there is a verbal description, mathematical models are hard for journalists to question. So it has been long recognised that modelling is a convenient way of avoiding scrutiny. However, the situation is so much worse when the practitioner has an agenda in mind and includes assumptions that are both dodgy and impenetrable.

References

Fisher B (2019a) Economic consequences of alternative Australian climate policy approaches, 11 March.

Fisher B (2019b) Economic consequences of Labor’s climate change action plan, 1 May.