##### Abstract

This paper shows that by omitting uncertainty when estimating the effects of a government spending shock on GDP could lead to biased estimates. The fiscal multiplier is how much an increase in government spending increases gross domestic product (GDP). There are two main theoretical views of looking at the economy: the neoclassical and the new Keynesian view. Neoclassical models predict multipliers that can be positive and negative depending on, among other things, the taxation system, because consumers determine their consumption based on their lifetime budget constraint. In Keynesian models the size of the multiplier depend on the consumer’s the marginal propensity to consume and these models predict positive multipliers.

The model used in this paper is a structural vector autoregression model (SVAR), this is a system of simultaneous equations. To estimate the effect of fiscal policy shock one must identify the system of equations. There are different ways to do this and there is wide disagreement among economist which approach is the best one. The first approach, and also the one used in this paper, is the approach introduced by Blanchard and Perotti in 2002, they use institutional information about the government to make the assumption that it takes more than one quarter for policymakers to react to changes in GDP. Then they use the Cholesky decomposition of matrices to identify the equation system and order government expenditure before output. This approach yields a positive multiplier, and finds that consumption rises in response to a government expenditure shock. This is in line with the New Keynesian view. Ramey and Shapiro (1998) use another approach, they use dates that are associated with large increases in military spending and see these as exogenous to the rest of the state of the economy to identify the equation system. Consumption actually falls in response to a government spending shock in this approach due to the negative effect taxes has on wealth. This is in line with the neoclassical view.

There are some challenges to these identification approaches; one of them is fiscal foresight. If a consumer foresees a change or a change in fiscal policy is announced consumers can alter their behavior before the change is implemented and the whole effect of a change in in consumption in response to a fiscal policy shock might not be captured. Blanchard and Perotti (2002) find that the response of output to a government spending shock was larger once they took account of the anticipation effects. Another problem is simultaneity bias created if a change in GDP affects expenditure. This will lead to inconsistent estimates. Other factors that influence the size of the multiplier is the composition and persistence of the shock, how monetary policy is conducted, the size of the automatic stabilizers, exchange rate system, the development level of the country, debt and the openness of the economy. In some rare cases a fiscal contraction has actually had expansionary effects, this is known as the “Austerity myth”. Sutherland (1997) uses a model and shows that a fiscal expansion will have Keynesian effects at lower levels of debt, but at high levels of debt, on the other hand, a fiscal expansion can have contractorary effects. This is because the consumers have finite lives and high levels of debt makes it very likely that a fiscal consolidation will take place in their lifetime. Keynes introduces the term “Animal spirits” to explain how changes in economic agents sentiment can lead to downturns and Bloom (2009) shows that uncertainty often jumps after large shocks like oil-price shocks and terrorist attacks. This leads to a drop in aggregate demand and employment in the short run, since firms’ pause/postpone their investment decisions. Higher uncertainty can cause consumers to increase their saving in order to prepare for a more uncertain future, this is known as “precautionary saving” (Carroll 1992). Also investment decisions are affected by higher uncertainty because it makes firms more cautious and therefore makes firms less responsive to stimulative policies (Bloom 2009).

The structural VAR used in this paper is based on the model in Caldara and Kamps (2008) and the Cholesky decomposition is used to identify the system of equations. As mentioned the identification approach used in this paper is the same one used in Blanchard and Perotti (2002), which turns the structural VAR into a recursive VAR. I have used MATLAB and performed the misspecification tests using the econometrics toolbox. The MATLAB code was obtained from the MATLAB file exchange, but minor changes have been made to the code. For the US I obtain a positive multiplier on impact and a negative cumulative multiplier for both models. In this case the multiplier is less negative when the volatility index measuring uncertainty is included in the model, which implies a that uncertainty has a negative bias on the effect a expenditure shock has on GDP. For the data from the United Kingdom I estimate a negative multiplier both on impact and cumulatively for both the models. However the multiplier is more negative when uncertainty is included in the model, by looking at the correlation between the volatility index and government expenditure I find a slightly negative correlation which means that uncertainty has a negative bias on the effect a expenditure shock has on GDP. In the case of Japan I estimate a positive multiplier in both of the models on impact and for the longer horizon. The multiplier is larger when uncertainty is included in the model, which implies that uncertainty has a negative bias on the effect a expenditure shock has on GDP. Norway is the only case where I find that uncertainty has a positive bias on the effect a expenditure shock has on GDP, but the multipliers are negative in both the models.

The results of the estimation show that uncertainty does lead to biased estimates. In the case of Norway I obtain an overvaluation of the multiplier due to uncertainty. But this is most likely because the measurement used to measure uncertainty is like the Norwegian economy correlated with the price of oil, which will be posititve for the Norwegian economy, but more of a disadvantage to the other countries in the sample. For the United States, United Kingdom and Japan uncertainty has a negative bias on the estimated multiplier, this makes the estimate undervalued.

Woodford (2010) shows how important monetary policy and the central bank’s response to changes in inflation is to determine the fiscal multiplier. Christiano et al. (2009) find that the fiscal policy has greater effect when the interest rate is low,this could explain the larger multiplier estimated for Japan because their Long-term interest rates have been very low compared to the interest rate in the other countries in the sample.

In this paper the omission of uncertainty led to negatively biased estimates for the multiplier in Japan, the UK and the US. I used the identification approach used by Blanchard and Perotti (2002). The identification approach used by Ramey and Shapiro (1998) estimates lower multipliers than Blanchard and Perotti (2002) but one might argue that during war uncertainty is probably also high. As far as I know they have not taken account of uncertainty in their model and this might be one of the causes for the lower multiplier. This would be consistent with my findings for Japan, the United States and United Kingdom, but not for Norway. Even though I obtain a negative bias for Japan, the US and the UK it is important to note that the effect of government expenditure is small also when uncertainty is controlled for.