RDP 2016-01: Measuring Economic Uncertainty and Its Effects 3. Stylised Facts about Uncertainty

This section uses the economic uncertainty index to understand the dynamics of economic uncertainty and document some stylised facts. In Section 3.1 I argue that the index passes the first sense test: it lines up with major events that might be expected to correspond to high uncertainty. In Section 3.2 I show that, although many of the major events from Section 3.1 are foreign, there is still a substantial amount of domestic variation in the index. Having established these properties, in Section 3.3 I document some statistical properties of the index. Finally, in Sections 3.4 to 3.6 I show that the index correlates with Australian recessions, elections and monetary policy decisions.

Appendix D contains some further stylised facts, including how the economic uncertainty index relates to measures of consumer or business sentiment.

3.1 Major Events

The index lines up with events that a priori might be expected to create economic uncertainty (Table 4; Figure 6). This fact is reassuring because it suggests the index is indeed capturing uncertainty.

Table 4: Major Events and Economic Uncertainty
Index, mean = 100 points
Date Event Uncertainty index Change from previous month
20 October 1987 (A) Black Monday (19 October in the United States) 190 +117
29 November 1990 (B) ‘Recession we had to have’ speech – approximate start of the Australian early 1990s recession 101 −19
November 1992 (C) Australian unemployment peaks at 11.1 per cent 155 −4
2 July 1997 (D) Baht floated – beginning of Asian financial crisis 73 +1
17 August 1998 (E) Russian Federation defaults 142 +2
23 September 1998 (F) Long-term Capital Management collapses 148 +5
11 September 2001 (G) 9/11 terrorist attacks 132 +47
20 March 2003 (H) US invasion of Iraq 108 +4
14 March 2008 (I) Bear Sterns rescued 120 +6
15 September 2008 (J) Lehman Brothers files for Chapter 11 bankruptcy protection 148 +9
23 April 2010 (K) Greece requests official financial assistance 102 +6
31 July 2011 (L) US debt ceiling stand-off – the debt ceiling is raised days before it would have been reached in August 132 +27
9 March 2012 (M) Greek debt restructuring becomes effective 100 −33

Sources: Author's calculations; Consensus Economics; Factiva; MSCI; Thomson Reuters

Figure 6: Major Events and Economic Uncertainty
Figure 6: Major Events and Economic Uncertainty

Notes: See note to Figure 5; see Table 4 for label definitions

Sources: Author's calculations; Consensus Economics; Factiva; MSCI; Thomson Reuters

Some of the events do not induce large in-month increases, but subsequent months have relatively large increases. Sometimes this is because the events occur late in the month – such as Greece requesting financial assistance in April 2010 – and so most of the increase occurs in the succeeding month. For some other events, uncertainty simply seems to respond more slowly – the 9/11 attacks induced a sizeable and rapid increase in uncertainty; the start of the Asian financial crisis induced a slower response. Some events correspond to decreases in the index because they resolve uncertainty; the Greek debt restructuring in March 2012 is a good example. Not all geopolitical events correlate with high estimated uncertainty; the index was not elevated around the US invasion of Iraq in March 2003.

3.2 Foreign and Domestic Uncertainty

As a small open economy, foreign economic shocks can have a large effect on Australia. Uncertainty is no exception – a number of major international events correspond to sizeable spikes in estimated uncertainty (Table 4, Section 3.1). However, regressing the economic policy uncertainty index for the United States from Baker et al (2015) on the Australian index accounts for less than a third of the overall variation in the Australian index (Figure 7). Foreign events such as the 9/11 attacks and the 2011 US debt ceiling stand-off are largely explained by the foreign component: the blue (Australian) and pink (US predicted) series line up. But most of the month-to-month variation in the index cannot be explained by foreign factors.

Figure 7: Foreign and Domestic Uncertainty
Figure 7: Foreign and Domestic Uncertainty

Notes: See note to Figure 5; the ‘US predicted’ series is from a regression of the Australian economic uncertainty index on the US economic policy uncertainty index from Baker et al (2015); ‘domestic component’ is the residual from that regression

Sources: Author's calculations; Baker et al (2015); Consensus Economics; Factiva; MSCI; Thomson Reuters

Foreign uncertainty seems to have been higher than domestic uncertainty since the global financial crisis. The two indices both increased rapidly during the early stages of the global financial crisis (top panel, Figure 7). However, the Australian index declined sooner. Indeed, the domestic component has never exceeded 30 points since the end of 2011.

3.3 Uncertainty Increases Faster than It Decreases

The business cycle is generally thought of as asymmetric – output grows in a relatively steady manner during expansions, but falls sharply during recessions. These types of dynamics are easily observed in the unemployment rate, but many have argued these dynamics are more general.[15]

The uncertainty index appears to be similar – increases in the economic uncertainty index tend to be larger than decreases. The distribution of changes in the index is positively skewed, with a skewness coefficient of 0.8. It is also fat-tailed, with excess kurtosis of 4.4. The distribution is significantly different from the normal distribution.[16]

Similarly, periods of high or low uncertainty tend to persist. The one-period autocorrelation of the index is 0.76 and the second-lag partial autocorrelation is 0.20. Further lags are not significantly different from zero. Despite the high degree of persistence, the economic uncertainty index is stationary – even with 12 lags, the p-value for the augmented Dickey-Fuller test is just 0.02.

Taken with Table 4, these facts suggest that the dynamics of uncertainty are often characterised by large spikes upward after major events, followed by slow reversion.

3.4 Recessions

Bloom (2014) shows that, for the United States, uncertainty is higher during recessions. In Australia, there have been few recessions during the period for which data exist for the economic uncertainty index. This makes it difficult to assess the relationship between uncertainty and recessions.

Nonetheless, there is some evidence that uncertainty is higher when unemployment is rising (Figure 8). The economic uncertainty index is on average 20 points higher in months where the unemployment rate is trending upwards than in months where it is trending downwards. In addition, trend employment growth and seasonally adjusted GDP growth are negatively correlated with the economic uncertainty index; however, the correlation is much weaker for GDP.

Figure 8: Unemployment and Economic Uncertainty
Figure 8: Unemployment and Economic Uncertainty

Notes: See note to Figure 5; shaded areas show periods where the unemployment rate is trending upwards; turning points are (somewhat arbitrarily) identified as the last minimum and the first maximum

Sources: ABS; Author's calculations; Consensus Economics; Factiva; MSCI; Thomson Reuters

3.5 Federal Elections

Federal elections are an opportunity to isolate a domestic event. Federal elections may elevate uncertainty because of uncertainty about the outcome of the election and the potential for changes to government policies. I test whether this is true by regressing the economic uncertainty index on dummy variables for near-election months. The coefficients represent the average in-month level of the index, relative to its remaining history. I have not controlled for any other relevant factors.[17]

Economic uncertainty is estimated to be higher than average before federal elections and in the month of the election (Figure 9). In the two months leading up to the election and the month of the election, the economic uncertainty index is 14 to 16 points higher than average, depending on the month.[18] This lines up with approximately when campaigning would begin to increase. As expected, estimated economic uncertainty declines following elections.

Figure 9: Federal Elections and Economic Uncertainty
Average of index in near-election months, relative to overall average
Figure 9: Federal Elections and Economic Uncertainty

Sources: Author's calculations; Consensus Economics; Factiva; MSCI; Thomson Reuters

However, Figure 9 masks a lot of dispersion: both the 2010 Federal election (which resulted in a hung parliament) and the 1998 Federal election coincided with in-month index levels of around 140; the 2004 election had an economic uncertainty index level of 91 points. The uncertainty around elections is probably related to how close the election result is. I do not take this into account here, but further work could examine whether closer elections – perhaps measured by poll results – align with higher economic uncertainty.

3.6 Monetary Policy Decisions

The RBA's monetary policy decisions are another set of domestic events which could be related to economic uncertainty. Changes in monetary policy that are not generally expected might be related to economic uncertainty either because: unexpected changes in monetary policy cause uncertainty (for instance, by making firms' financing decisions less certain); or, when economic uncertainty is higher, it might be harder for market participants and commentators to predict the path of monetary policy (and other economic aggregates).

I measure ‘monetary policy surprises’ by the absolute difference between the market-implied cash rate the day before a cash rate decision and the subsequent decision. These surprises capture both unexpected movements and anticipated movements that occur earlier or later than expected.

The economic uncertainty index is indeed correlated with ‘monetary policy surprises’ (Figure 10). The two series have a correlation coefficient of 0.49, which is statistically significant at the 1 per cent level, despite the fact that I only have market-implied cash rate data going back to February 2005. Of course, these correlations cannot disentangle which way causality flows between uncertainty and monetary policy.

Figure 10: ‘Monetary Policy Surprises’ and Economic Uncertainty
Figure 10: ‘Monetary Policy Surprises’ and Economic Uncertainty

Notes: See note to Figure 5; ‘monetary policy surprises’ are measured by the absolute difference between the overnight indexed swap-implied cash rate at market close the day before the RBA Board meeting and the actual cash rate decision

Sources: Author's calculations; Consensus Economics; Factiva; MSCI; RBA; Thomson Reuters


For example, Burns and Mitchell (1946), Hamilton (1989) and Morley and Piger (2012) among many others. [15]

Based on a Shapiro-Wilk test. Additionally, the distribution differs from a normal distribution on both skewness and kurtosis at the 1 per cent level, based on a modified D'Agostino K-squared test. [16]

I also tried a specification that attempted to control for foreign-based uncertainty by using the residuals from the regression of the Australian index on the US index (see the bottom panel of Figure 7). This did not materially alter the results. [17]

In part because there are only 10 elections in the sample, the confidence interval around these estimated effects is large. [18]