Fiscal developments and financial stress: a threshold VAR analysis

See for example Arghyrou (2014) for an attempt to provide an assessment of the sovereign debt crisis. The feedback from government debt to financial instability became relevant for numerous countries in the EU that were hit by the financial meltdown seriously like Ireland or Spain, countries with comparably low debt/GDP ratios before 2008. The channel how the perceived un-sustainability can worsen the outlook for the prospective GDP growth can be described as follows: when the debt sustainability is concerned, financial markets may refuse to buy new government debt or roll-over the old one, while transactions in the secondary market may also become less frequent. The inability to sell government bonds in the market reduces the liquidity of the debt and weakens the balance sheet of the banks and of other financial institutions that hold government debt.

The multipliers of specific instruments of fiscal policy are surveyed in Gerchert and Rannenberg (2014).

An overview of the effectiveness of fiscal policy and evaluation of fiscal multipliers in several VAR models is provided in Van Brusselen (2010) and Hebous (2011). The role of the different identification schemes for the estimated multipliers has been investigated by Caldara and Kamps (2008), with alternative identifications: SVAR following Blanchard and Perotti (2002) and Perotti (2005); recursive identification scheme with Choleski decomposition; sign-restriction approach following Uhlig (2005) and Mountford and Uhlig (2009); narrative approach for the USA, along the lines of Ramey and Shapiro (1999). The results by Caldara and Kamps (2008) were supported by a meta-analysis of fiscal multipliers by Gechert and Will (2012).

See, notably for fiscal consolidations, Alesina and Ardagna (1998), Afonso and Jalles (2014), and Afonso and Martins (2016). In particular, policy makers in Europe were strongly influenced by the hypothesis of expansionary fiscal consolidations. In addition, Giavazzi and Pagano (1990) pointed that the costs of the debt were so high that decreasing the debt via fiscal consolidation allowed rise of private consumption through improvements in expectations about taxation in the future. The so-called non-Keynesian effects dominated the discussion about the adequate response to the Great Recession in Europe following a contribution by Alesina (2010). However, the recent evidence shows that fiscal contractions are likely to be recessionary and the non-Keynesian effects are exceptional (see for example Guajardo et al. (2014).

We used the WinRATS code provided by Nathan Balke, which we modified for our purposes.

Cardarelli et al. (2011) have shown that the components of the FSI are relatively uncorrelated and, importantly, adding different variable does not change the resulting path of the FSI significantly. Regarding the exchange rate component, we do not observe, for Germany and Italy, significant changes around the adoption of the euro in 1999. For Italy, some volatility can be seen after the exiting of the Italian Lira and of the British Pound from the European Exchange Rate Mechanism on September 1992. The version of the FSI in Cardarelli et al. (2011) is constructed by taking the average of the components after adjusting for the sample mean and standardizing by the sample standard deviation. Then, the index is rebased so that it ranges from 0 to 100. Episodes of high stress are identified as those periods when the FSI is more than one standard deviation above its trend determined by the Hodrick–Prescott filter.

Balakrishnan et al. (2009) suggest the value of one for the FSI to distinguish the periods of high and low stress. Their judgment is based on the experience that such identification of stress periods mimics well the historical episodes of financial instability.

The threshold variable shall be stationary to assure alternation of the regimes. Stationarity of both, the FSI and its moving average has been confirmed by the ADF test.

One can conjecture that these results for Italy and the UK might reflect the fact that the slowdown in economic activity was driven by fiscal consolidation and not primarily by financial instability.

For details see “Appendix 1”.

Other authors include just one variable representing fiscal policy in threshold VAR to save the degrees of freedom as well. For example, Ferraresi et al. (2014) consider the real government consumption and gross investment (GCE) and Ramey and Zubairy (2014) focuses on the effects of military expenditures.

The change in the debt ratio and the budget balance ratio are rather correlated for the countries in our analysis.

It can be also noted that extensive sensitivity tests have been performed also by Ramey and Zubairy (2014). They re-estimate their model with potential GDP in the denominator of the government spending variable finding that this change has little effects for the results.

Note that the impulse responses of output were divided by the initial size of the fiscal shock to normalize sizes of the shocks to 1% of GDP. The values of the impulse responses equal to \(\Delta Y_{t+k} / \Delta D_{t}\) where t is time of shock and k horizon of the impulse response, thus correspond to multipliers at horizon k.

Although we have cleaned the data for the impact of German unification on the level of GDP and for the transfer of the debt of social institutions of the former GDR to the federal debt in the mid of 1990s, there were numerous other large spending programs in the former GDR financed either through new taxes or deficits. These expenditures include: costs of privatization, social benefits, investment into infrastructure etc.

When we multiply the size of the rise in FSI in the Great Recession by a size of the peak of the impulse response of debt ratio to a shock in financial stress, we obtain the following increases of the debt ratio: USA 8.7% of GDP, UK 10.7% of GDP, Germany 1.5% of GDP, Italy 2% of GDP. With the simulated impulse responses for the year 2000, the increase in the debt ratio is about twice as large.

The BIC selects 1 lag for models with y-o-y changes and 2 lags for the model in first differences. We also tested whether the nonlinearity is significant in these models, and the nonlinearity has been supported by the tests again. The multipliers were retrieved from the impulse responses via multiplication by the average GDP/debt and GDP/GCE ratio.

There are, unfortunately, no equivalent data to the GCE series also for other countries in our sample that would be available for a sample from the 1980s and on quarterly frequency Table 7.