16 Aug 2012 by Jim Fickett.
In looking for a link between higher government spending and economic growth, one approach is entirely empirical, as exemplified by the scatter plots mentioned or shown in previous posts of this series. The other approach involves models of the economy. The empirical approach is the closest to reality but, by design, cannot address causation. One can perturb individual variables of a model, and see what happens, allowing one to query causation. The faith of many economists in fiscal stimulus probably rests on results from highly complex models, however these models may do little more than transform faith in various theoretical assumptions and constructs into faith in stimulus. Results from one very simple model, consisting of little more than a few empirical correlations (and hence closely tied to reality), suggests that fiscal stimulus is not typically useful. It is quite likely that governments are really out of options for a quick fix, and we will see more years of sluggish growth.
To give some feeling for typical econometric models in use today, consider the widely cited Area-Wide Model of the European Central Bank. The AWM treats the euro area as a single economy, and models the behavior of this economy using 89 equations. The very first equation illustrates why I am hesitant to accept results based on such models; this equation calculates potential economic output – the level of GDP that theoretically could be produced on a sustained basis – as a function of
I doubt very much that anyone really knows what level of employment is consistent with stable inflation, or what the potential level of economic output might be. It is telling that in the 63-page paper describing the AWM, forecasting with the model is mentioned many times, but there is no mention of back-testing to see if forecasts using past data actually would have given the right answers.
Still, some models are useful. In general, if a model has only a few variables, and the calibration of the model is robust to the use of slightly different assumptions and calibration data sets, then one can at least treat results of the model as likely to be related to reality.
A refreshingly simple model may be found in the work of Roberto Perotti, described in the paper Estimating the effects of fiscal policy in OECD countries. In the introduction to this paper, Perotti writes,
perfectly reasonable economists can and do disagree even on the basic qualitative effects of fiscal policy. For instance, neoclassical models predict that private consumption should fall following a positive shock to government consumption, while keynesian and some (though not all) neokeynesian models predict the opposite. It would seem that empirical evidence is what is needed to make further progress.
Indeed – look at the actual data! A man after my own heart.
Perotti sets up a very simple model involving only
These are combined in a simple vector autoregression model; you will not be far off if you think of this model as just relating the value of each of these variables to all the others by their correlation coefficients.
Perotti studies the effects of fiscal policy in 5 OECD countries: the US, West Germany, the UK, Canada, and Australia. When he calibrates the model using historical data from each country, he finds that data from before 1980 yields quite a different model than data from after 1980, and so he ends up with 10 models – a pre-1980 model for each country and a post-1980 model for each country. (The 1980 date is approximate, and Perotti does check that his main conclusions are robust to slight changes in the data and model.) For each of these 10 models, he then asks what the effects are of a spike on government spending equal to 1% of GDP.
Here are the key results. The five countries are across the top, with two delay periods considered for each – 4 quarters and 12 quarters. Six facets of the model's response to the spending spike are given in panels A to F of the table; panel E, giving the response of GDP, is the one of interest for this post. On the left, “S1” is the pre-1980 period, and “S2” is the post 1980 period. So the first number in panel E says that in the pre-1980 model for the US, a spending increase of 1% of GDP results, after 4 quarters, in a cumulative increase of 1.13% in GDP.
(Click for larger image.)
Here is the concise summary of results for the post-1980 period. When government spending is increased by 1% of GDP,
Perotti also looks at the effect on growth of tax cuts. His overall summary of both kinds of stimulus is:
In S2 [the post 1980 period], a spending - induced deficit stimulates output only in the US and in Australia; a tax - induced deficits, only in Canada. In most other cases, both shocks have negative effects.
There is a large economics literature relevant to whether a fiscal stimulus causes the economy to grow. I have read only a few papers and am not prepared to form a strong opinion for myself. What I can say, though, is that there is clearly room for disagreement among the experts, and no easy proof that fiscal stimulus is effective. Thus the common assumption, in policy discussions, that the only question is how much stimulus we can afford, is quite wrong.
As an investor, I do come down on the side of doubting the effectiveness of fiscal stimulus, particularly in cases of high national debt. This is one more reason, among many, to expect that a “new normal” of sluggish growth will be with us for a long time.