Treasury says yes. In the budget papers they produce a graph showing a nice upward sloping relationship between the size of the stimulus in 2009 relative to IMF forecasting errors.
Those countries that enacted large and timely fiscal stimulus packages, including China, Korea and Australia, performed much better than expected. Those countries with smaller packages, such as the US, Germany, Canada and France, tended to perform broadly in line with expectations. The relationship shown is highly statistically significant, with a t?statistic on the slope coefficient of 3.3.
Before we go too far into the analysis, I should point out that this confuses correlation and causation. In addition the logic could be reversed. You could easily argue that some economies got paniced into enacting large stimulus packages by IMF forecasts that turned out to be very wrong.
Possum verifies the econometric result and comes to the same conclusion as does Treasury.
The size of stimulus packages mattered – the international evidence is in.
The Coalition needs to be questioned about its economic viewpoints – viewpoints which are far from mainstream economics, existing on the very fringes of economic debate and which are completely at odds, completely and utterly at odds, with the international empirical evidence.
You know what an argument without evidence is called?
Making shit up.
Data snooping is also ‘making shit up’. That is exactly what that graph is – data snooping. Treasury take their data from the IMF. They use the size of stimulus as a percentage of 2009 GDP (for 19 of the G20 economies – the EU is a member in its own right) from here (pg.36) and then calculate the forecast error by subtracting the 2009 forecast GDP growth from the 2009 actual GDP growth. I do have a minor quibble – some of the 2009 actuals are still forecasts (Possum looks like he has substituted an actual for Australia in his regressions), in the analysis below I use the IMF data as downloaded. Like Possum and the Treasury I use the absolute value of the stimulus size and not the negative figure.
The problem with the Treasury analysis is that it only makes use of 11 observations, when they could have used 19 observations. The table below shows the data they could have used relative to the data they did use.
There is no obvious reason why those countries should have been excluded. The Treasury sample isn’t just advanced economies and isn’t just resource rich economies. What is the effect of using the truncated sample? First I show the graph as per the Treasury sample and then I show the graph as per the full sample.
This is the graph making use of the full dataset.
The slope coefficients are positive in both cases. The t-stat for the slope coefficient for the full data set, however, is 0.50 well below the generally accepted levels for a t-stat to indicate statistical significance. In layman terms the slope coefficient is not statistically significantly different from zero and we cannot conclude that there is a relationship between the size of stimulus packages and the forecast error. To emphasise the difference between the results for the Treasury analysis on the truncated sample and the full sample, I reproduce the two lines of best fit together in the graph below.
What I’ve done is take the regression equations from the two graphs and then substituted x values into them from 0.5 to 4 at 0.5 intervals. I then plotted the lines. There is a huge difference between the two – especially for the three economies (Australia, China and Korea) that Treasury nominate as having had early and large stimulus packages.
So far the discussion has been about the potential for outliers to confound the analysis and so on. But before we get to that, we should be looking at the data sample. The IMF reported data for 19 of the G20 economies, and Treasury have used a sub-sample of those economies in their analysis. If there is a good reason for having done so, they need to tell us.
Update: Lurkers email to ask why this is being released today – these are the workings for Hey, what did I miss? If you don’t already have one, get an email subscription.