«Jacopo Ponticelli Hans-Joachim Voth UPF ICREA/UPF and CREI Abstract: Does fiscal consolidation lead to social unrest? From the end of the Weimar ...»
Using almost a century of data allows us to include some extreme observations. For example, Austria and Germany saw major output declines in 1945 and 1946, respectively. The biggest reduction in governments spending in our data occurred in Poland, in 1982; the second-largest, in Finland, in 1947. The start of war is often associated with big increases in expenditure. The record-holder in our dataset is Hungary in 1940, with an increase of over 30 percent.
Table 1: Descriptive statistics, main variables To obtain a single measure of instability, we calculate CHAOS by taking the sum of the number of assassinations, demonstrations, riots, general strikes, and attempted revolutions. While a crude way of aggregating indicators, it turns out to be powerful.8 One alternative is the weighted conflict indicator (wci), as compiled by Banks (2010). It gives fixed weights determined to different forms of unrest: Demonstrations have a weight of 200, while political assassinations have a weight of 24.
In the robustness section, we show that alternative methods of reducing data complexity such as principal components analysis do not change our results.
For CHAOS, the average country in our sample registers 1.5 incidents per year. Instability was not constant over time. The maximum is higher – Italy in 1947 saw a total of 38 incidents, including 7 general strikes, 19 riots, and 9 antigovernment demonstrations. Figure 1 gives an overview of the evolution over time, plotting the average of CHAOS as well as the maximum number of incidents observed. While there is no clear-cut pattern over time, some features emerge. The interwar period showed relatively high levels of unrest, with an average of 2 incidents per year, compared to 1.4 in the post-war period. The immediate post-World War II period, and the period form 1968 to 1994 also show unusually high levels of unrest.
Comparatively speaking, the years since 1994 have been unusually tranquil (average CHAOS = 0.78) number of incidents (CHAOS)
Figure 1: CHAOS over time The short-term data on unrest is from the European Protest and Coercion Database (EPCD) developed by Francisco (2000). The EPCD codes daily data on all reported protest events occurred in 28 European countries between 1980 and 1995. The data is constructed using the full-text reports from more than 400 newspapers in the LexisNexis database. We restrict our attention to the same types of protest events covered in the long-term data: riots, demonstrations, political assassinations, general strikes, and attempted revolutions.9 The main advantage of the EPCD over the Arthur Banks’ database is that the former records the issue behind each protest, allowing us to test the relationship between austerity and unrest in a very precise way, even if only for a small subset of the overall dataset.
There are relatively few protests that are caused by austerity measures. At the same time, when they happen, they involve a large number of participants – by far the largest number of protesters of any category, as Table 2 illustrates. These protests tend to be relatively peaceful, with few protesters arrested, injured or killed, and relatively few members of the security forces involved.
Table 2: Unrest in the EPCD sample, 1980-95.
In compiling information on expenditure and the budget balance data, we need to trade off the accuracy of information against availability over a long time span. For the 1919-2009 dataset, we rely on standard data sources on the central government revenue and expenditure relative to GDP (Mitchell 2007) for the years 1919 to 1970, augmented by data from the OECD (2010) for the period thereafter.
Expenditure changes will serve as the main explanatory variable. Figure 2 graphs changes in expenditure/GDP from one year to the next. The distribution is almost symmetric around the mean, with similar numbers of country-years witnessing expenditure increases and declines (807 vs 685). In an average year and country over the period, central government expenditure relative to GDP rose by 0.3%. The vast majority of observations falls between increases and decreases of 5%, with a few We only consider protest events whose number of participants is above 100 for riots and demonstrations and above 1000 for general strikes (no threshold is used for assassinations and attempted revolutions). These are the same threshold used in the Arthur Banks database.
outliers in the tails of the distribution (typically driven by the beginning and end of wars).
Figure 2: Expenditure changes/GDP, 1919-2009, all countries In addition, we use the data by Alesina and Ardagna (2010) for the cyclically-adjusted budget balance.10 This has the advantage of correcting the budget position for changes in interest payments and for the immediate effect of the economic cycle, which drives both expenditure and revenue without any additional policy decision being taken. For a subsample of the data (1978-2009, 17 countries), we also use data by Devries et al.
(2011). These authors examine in detail the policy changes that led to changes in a country’s fiscal stance. Only expenditure cuts or revenue increases motivated by a decision to press ahead with fiscal consolidation are considered.11 Overall, Devries et al. (2011) find 173 periods of fiscal policy adjustment, As a first pass at the data, we repeat the exercise in Figure 1 for output growth (Figure 4). We subdivide the sample into terciles, and examine how much the incidence of various indicators of unrest declines as growth accelerates. For the summary indicator (CHAOS), there are a little more than 2 incidents when growth is in the lowest tercile. This falls to 1.3-1.5 incidents as growth accelerates. There is also a clear pattern of decline for demonstrations and for assassinations. In the case of Alesina and Ardagna use the method of Blanchard (1990b).
The approach is similar to the “narrative approach” pioneered by Romer and Romer (1989).
riots, the differences are smaller overall, whereas in the case of general strikes, there seems to be little pattern at all. Based on a first, visual inspection of the data, it seems that the link between budget cuts and unrest is clearer than the one with growth.
Figure 4: Frequency of incidents and economic growth Next, we examine the correlation structure of our data in Table 3. Assassinations, general strikes, riots, revolutions and demonstrations are all positively and significantly correlated with each other. This supports our assumption that they reflect a broader underlying pattern of social instability and unrest. CHAOS is also positively correlated with the weighted conflict index (wci). Finally, Table 3 suggests that higher levels of expenditure and faster growth are associated with less unrest. The simple correlation of CHAOS with changes in the budget balance is positive and significant.
Higher taxes and lower expenditure are associated with more unrest, but the relationship is not significant.
Table 3: Correlation matrix, main variables (significance levels in parentheses) In the case of output changes, the coefficient is negative, but insignificant (table 3).
The simple correlations suggest that these co-movements do not extend to all indicators of unrest equally – riots, revolutions, and demonstrations decline as expenditure rises, but assassinations and strikes seem – at a first pass – uncorrelated.
Similarly, output growth seems to correlate negatively with assassinations, riots, revolutions, and demonstrations, but not with strikes. Next, we examine the connection between budget position, expenditure, and unrest more systematically.
The graphical evidence in Figures 1 and 4 suggests a link from “hard times” – low growth and budget cut-backs – to unrest. Next, we examine if there is a systematic relationship between budget measures and social instability. In this section we also address the issue of causality, while in the next section we will test the robustness of our results.
A. Baseline Results
We estimate panel regressions of the type:
I it i t Bit X 'it it (1) where Iit denotes the level of instability in country i at time t, B is an indicator of the change in the budget position, is a country-specific intercept, is a time-specific dummy, and X’ is a vector of control variables.
We use CHAOS as the dependent variable in our baseline specification, and test the robustness of findings to alternative specifications later. Table 4 gives the main results. Under OLS with fixed effects and year-dummies, we find that expenditure increases reduce instability in a powerful way (column 1). A one standard-deviation increase in expenditure cuts the number of incidents (CHAOS) by
0.4 per year and country. Tax increases have a positive sign, but the effect is not significant at standard levels of rejection (column 2). It is also small – a one standard deviation rise in the tax/GDP ratio increases unrest by less than 0.01 events. Overall, we find that improvements in the budget balance raise the level of unrest (column 3).
As the results in columns (1) and (2) make clear, this reflects the impact of expenditure cuts, and not of tax increases.
CHAOS is a count variable. Hence, the use of OLS may not be appropriate.
Columns (4)-(6) give the results for Poisson Quasi-Maximum Likelihood estimation, with fixed effects. We find the same pattern as before, with strong effects for expenditure cuts and much weaker ones for tax increases.12 We also experimented with using negative binomial regressions, but results were largely unchanged.
Table 4: Baseline results Which component of CHAOS is responsible for the significant predictive power of budget cuts? In Table 5, we use the same specification as in Table 4 under Poisson QML, looking at the effect of expenditure cuts on each of the components of the aggregate indicator of instability – general strikes, demonstrations, riots, assassinations, and attempted revolutions. Out of the five outcome variables, four show the expected sign, and all of them are statistically significant. The only variable that does not show a large, significant coefficient is general strikes. On average, years with expenditure increases showed fewer general strikes, but there are numerous general strikes that are not an immediate reaction to economic conditions and budget measures (such as, for example, the 1926 general strike in Britain). For the other variables, the coefficients are large, indicating that austerity measures coincide with significant increases in demonstrations, attempted revolutions, riots, and assassinations.
In all specifications, the effect of GDP growth on unrest is negative. In contrast to the results for expenditure changes, the effect is not tightly estimated, except in the case of demonstrations, when it is also large – every 1% increase in GDP cuts the number of demonstrations by close to 0.4 events.
Table 5: Fiscal Adjustment and CHAOS by component
Table 6 takes this analysis one step further, by breaking the period 1919-2009 into four sub-periods. We distinguish the interwar period from the period of immediate post-World War II reconstruction, the period of slowing growth into the 1980s, as well as the years after the fall of the Berlin Wall after 1989. On the whole, we find the same pattern as in the sample as a whole, with the exception of the last two decades.
The effect of changes in budget expenditure on unrest is strongest in the tumultuous interwar years, when the estimated coefficient is fifty percent larger than in the sample as a whole. The effect of GDP growth is negative, but not tightly estimated. In the years after 1945, the inverse relationship between expenditure and unrest remains.
Strikingly, however, more growth now appears to lead to more unrest. While it is difficult to test for the causes of this reversal exactly, it seems that high rates of output growth may have encouraged worker militancy more generally. At a time when many countries reached full employment, this effect seems to have become dominant. The normal pattern of GDP growth reducing unrest reasserts itself after 1965, when there is also still a clear negative effect of higher government expenditure.
The fall of the Berlin wall saw the spread of Western-style democracy eastwards. The overall connection between austerity and social instability now changes sign, and becomes in insignificant. This suggests to us that non-economic causes became a dominant feature of the period. Below, we examine the issue in more detail with the help of a dataset that allows us to look at the motive of each demonstration.
Table 6: Results by sub-period and sub-sample
B. Causality The obvious challenge in interpreting (1) is the potential for omitted variable problems. It is possible that the economic cycle is simultaneously driving both unrest and the need for budget cuts. Above, we already control for GDP growth rates, and our main finding remains unaffected. However, the omitted variable problem would only be solved if we measured the effect of economic output on instability perfectly.
Since this is unlikely, we present a different add two type of analysis. We use a related dataset that offers detailed information, for a shorter time period, on the causes behind each unrest event. This allows us to demonstrate the connection between social instability and expenditure cuts more directly.
As described in the data section, the EPCD’s dataset allows us to pin down the main motive behind each public demonstration. We examine if the public assemblies that are motivated by anti-austerity sentiment – as determined by the newspaper records in Lexis-Nexis – are significantly affected by actual changes in fiscal policy.
Our approach here is similar to what has been called the “narrative approach” (C.D.