RDP 2016-01: Measuring Economic Uncertainty and Its Effects 4. The Effects of Uncertainty on the Australian Economy

In this section, I examine whether and how uncertainty matters for the Australian economy. I do so by estimating two vector autoregressions (VARs) – one at monthly frequency and one at quarterly frequency.

Theory predicts that heightened uncertainty should delay decisions that are costly to reverse – such as hiring and investment. Once uncertainty subsides, businesses should then catch up on these delayed decisions, leading to overshooting. This is the ‘real options’ channel of uncertainty (Bernanke 1983; Dixit and Pindyck 1994; Bloom 2009).^[19] If the theory is valid, we should see initial falls in investment and employment growth in response to an unexpected increase in uncertainty, followed by growth, then overshooting. The precautionary savings channel of uncertainty predicts that the household saving ratio should rise and that consumption growth – particularly discretionary consumption – should fall following a rise in uncertainty (Kimball 1990; Carrol 1997).

4.1 Monthly VAR

In this section I test the real options channel of uncertainty on employment. Because labour force data directly address the real options channel for employment and are available monthly, I use a monthly VAR. This allows me to exploit the monthly variation in the uncertainty index. In addition to uncertainty and employment, I include three other relevant economic series that are also available monthly: retail sales growth; the cash rate; and ANZ-Roy Morgan consumer confidence.^[20]

I use one lag in the VAR. The results appear to be robust to using one, two or three lags; I choose to use one for parsimony. Appendix E.1 shows results from alternative specifications.^[21]

I use a Cholesky decomposition with the economic uncertainty index ordered last. This assumption imposes the restriction that shocks to the other economic variables affect the economic uncertainty index contemporaneously, but uncertainty only affects other variables with a lag. This is a conservative identification assumption. By ordering the economic uncertainty index last, the identified uncertainty shocks are purged of any predictable component due to any of the other variables. For sentiment in particular, this can be material. As a result, this assumption will tend to reduce the size of the estimated responses compared to alternative orderings. For this reason, the estimates can be thought of as representing a lower bound.

Much of the empirical literature takes the opposite approach and orders uncertainty first. Because of this, I also present estimates that have the economic uncertainty index ordered first (Appendix E.2). The effects of uncertainty are larger with this identification assumption, but the difference is generally not large and the qualitative results are unchanged. Since the more-conservative assumptions are enough to show that uncertainty matters, I use these in the main text.

The estimated effects of a one standard deviation (30 point) shock to the economic uncertainty index on economic activity are shown in Figure 11. The key result is that employment growth is persistently lower. However, the peak effect is only modestly sized – annualised monthly employment growth declines by about one-sixth of a percentage point after four months. For comparison, average annualised monthly employment growth over the period is 1.8 per cent. The response is statistically different from zero for four months (at the 5 per cent level, between six and nine months after the shock), but the confidence intervals are wide and an initial increase in growth is comfortably within the confidence interval. This response provides support for the real options channel. In contrast to real options theory, I find no evidence of overshooting as firms catch up on hiring decisions that were delayed by uncertainty; however, there is a deal of uncertainty around more-distant estimates, and I cannot reject an increase in growth after ten months.

Retail sales growth decreases, but the magnitude is small – annualised monthly growth is reduced by about one-sixth of a percentage point, compared to average annualised monthly growth of about 5 per cent. The confidence intervals are wide and include a large mass above zero. This response therefore provides little evidence in support of a precautionary savings channel, which predicts that consumption growth should be lower. However, retail sales growth is not a complete measure of consumption and it is also a relatively poor measure of discretionary consumption, which the precautionary savings channel predicts should be most affected.^[22] I examine the precautionary savings channel with better-suited quarterly data in Section 4.2. Consequently, I place little weight on the precautionary savings evidence from the monthly VAR.

Uncertainty shocks are estimated to have a small effect on consumer confidence. Following an uncertainty shock, consumer confidence decreases by 1.2 points after three months, although it is slightly positive after a little over a year, before returning to baseline. This initial decrease equates to a little less than a tenth of a standard deviation. The small magnitude of this change is reassuring – it suggests that the responses do reflect uncertainty shocks, not combined sentiment and uncertainty shocks.

Uncertainty shocks have a noticeable effect on the level of the cash rate, which decreases by a little less than a quarter of a percentage point within a year after the shock. Monetary policy is potentially responding to the likely effects of uncertainty on other relevant macroeconomic variables. Such a response of the cash rate might help mitigate the effects of uncertainty on employment somewhat.

4.1.1 Does foreign or domestic uncertainty matter?

As noted in Section 3.2, foreign uncertainty appears to be an important source of uncertainty for Australia. As a small open economy, this is unsurprising. In this light, it is possible that foreign uncertainty shocks affect economic outcomes differently than domestic shocks.

I test this hypothesis by using the US economic policy uncertainty index from Baker et al (2015) to separate foreign and domestic uncertainty shocks (Figure 12).^[23] I do so by ordering the US index ahead of the Australian index in the monthly VAR and restricting the response of the US index such that it does not respond to any of the Australian variables; all Australian variables are able to respond to shocks to the US index.^[24]

It appears that domestic and foreign shocks have similar effects (Figure 12). Although the foreign shocks appear to have a marginally larger effect, these responses are not precisely estimated and the confidence intervals are wide. In sum, these results suggest that foreign and domestic uncertainty shocks affect Australian economic outcomes similarly. It is uncertainty that matters, not the source of uncertainty. In the quarterly VAR I do not distinguish between foreign and domestic uncertainty shocks.

4.2 Quarterly VAR

I test for evidence of the precautionary savings and real options (investment) channels using a quarterly VAR that incorporates eight data series: growth in machinery and equipment (M&E) investment; growth in household final consumption expenditure on durable goods; growth in employment; the household saving ratio; changes in the terms of trade; the quarterly average of the cash rate; the quarterly average of ANZ-Roy Morgan consumer sentiment; and the quarterly average of the economic uncertainty index.

For the reasons discussed in Section 4.1, I again adopt a conservative identification strategy by using a Cholesky decomposition with the economic uncertainty index ordered last. As with the monthly VAR, Appendix E.4 shows that ordering the economic uncertainty index first increases the responses, but the difference is not substantial and the overall story is unchanged.

I estimate a VAR with two lags.^[25] The results are robust to other lag specifications and orderings, with the exception of using one lag with the economic uncertainty index ordered last (Appendix E.5).

Consistent with the predictions of the real options channel of uncertainty, M&E investment growth decreases and remains persistently weaker following a one standard deviation uncertainty shock (Figure 13); however, the confidence intervals are wide and cover a large mass above zero. The peak magnitude of the reduction (after three quarters) in annualised quarterly growth is a little less than 4 percentage points; this response compares to average annualised quarterly M&E investment growth of a little less than 7 per cent.^[26] However, as with the monthly results from Figure 11, I find no evidence of overshooting.

The initial increase in growth in the quarter following the shock is surprising, and is followed by a sharp reversal and more persistent reduction in growth. Ordering the economic uncertainty index ahead of consumer sentiment (or first) largely eliminates this initial spike (Appendix E).

The results support the precautionary savings channel: the household saving ratio rises by just more than half a percentage point and remains persistently elevated. This increase is not insubstantial relative to the average household saving ratio of around 5½ per cent over the sample period. Also consistent with the precautionary savings channel of uncertainty, annualised quarterly growth in the consumption of durable goods falls by as much as 1.3 percentage points.^[27] This reduction compares to average annualised quarterly growth of a bit more than 3 per cent. As with the saving ratio, this is a not insubstantial effect and lends reasonable support to the precautionary savings channel.

In terms of the other responses from the quarterly VAR, employment growth appears broadly similar to the response in Figure 11. Reassuringly, the reduction in (annualised quarterly) growth between two to eight quarters (inclusive) is between about 0.15 to 0.20 percentage points, which is a little larger than the roughly one-sixth of a percentage point response estimated at monthly frequency around months two through six. The response in the quarterly VAR is also much more persistent – in the monthly VAR the effect has largely dissipated after two years, but in the quarterly VAR the response is estimated to be a little over 0.1 percentage points at the same point. As noted for investment growth, the initial increase in employment growth is surprising in light of both theory and the subsequent persistent reduction. The response of consumer confidence and the cash rate are similar to those estimated in the monthly VAR, although a little smaller. The terms of trade barely respond to an uncertainty shock. For more detail see Appendix E.3.

4.3 Comparison to Other Literature

There is a large empirical literature that attempts to assess the effects of uncertainty shocks; the vast majority of it focuses on the United States.^[28] My estimates of the effects of uncertainty are generally a little smaller.

Based on a similar aggregate-level VAR framework, Bloom (2009) finds that uncertainty shocks lead to a 1 per cent decrease in industrial production and about a 0.7 per cent decrease in employment. In a similar vein, Caggiano et al (2014) use a smooth transition VAR – which allows them to differentiate between recessions and expansions – and find that an uncertainty shock raises the unemployment rate about 0.17 percentage points four quarters after the shock. During a recession, the effect is larger – about 0.36 percentage points. They also find the level of investment falls about 3 per cent below trend during a recession in response to an uncertainty shock. They argue that distinguishing between uncertainty shocks in recessionary and non-recessionary phases is important.^[29]

Bloom et al (2013) use a calibrated macroeconomic model to study the effects of uncertainty on economic activity. They find that uncertainty shocks decrease the level of GDP by about 2 per cent below trend, and hours worked by about 2.5 per cent. The decrease they find is larger, occurs much faster, and is less persistent than what I estimate for employment. In a similar vein, Leduc and Liu (2015) use a macroeconomic model with search frictions and nominal rigidities. They find that an uncertainty shock increases the unemployment rate by about 0.15 percentage points at peak (after about 18 months). This result is a little larger than what my estimates for employment growth imply for the unemployment rate. In contrast, Bachmann and Bayer (2011) argue that uncertainty shocks are only a small driver of fluctuations in economic activity.

Carriero et al (2015) argue that the measurement error inherent in proxies for uncertainty can substantially attenuate the estimated impulse responses, but using the uncertainty proxy as an instrument eliminates the bias. They apply their method to the dataset from Bloom (2009) and find substantially larger effects from uncertainty than Bloom. These results suggest that my estimates may understate the effect of uncertainty.

In terms of Australian studies, Tran (2014) uses firm-level data on investment and both firm- and aggregate-level measures of uncertainty to assess how uncertainty affects investment. The results suggest that uncertainty weighs on investment, and that firm-specific uncertainty is more relevant than aggregate-level uncertainty.

Other Australian macroeconomic VARs have tended to focus on identifying monetary policy shocks. These include Berkelmans (2005), Lawson and Rees (2008) and Dungey and Pagan (2009) among others. A possible extension of my work is to examine how uncertainty might affect the results of those papers, particularly given that Caggiano et al (2014) find asymmetric effects of uncertainty during recessions.

Footnotes

The risk premia channel also suggests that uncertainty should reduce investment by raising the cost of capital for firms. I do not address the risk premia channel because I do not have the data necessary to disentangle it from the real options channel. [19]

I use consumer confidence because it has a longer monthly time series than business confidence. In Appendix E.1 I show that the results are robust to using other measures of consumer and business sentiment. [20]

The Akaike Information Criteria (AIC) recommends three lags; the Schwarz Bayesian Information Criteria (BIC) recommends one. [21]

In particular, the retail sales measure is about two-thirds non-durable consumption. This weight is higher than the national accounts measure of goods consumption. In Section 4.2 I use a measure of durable consumption from the quarterly national accounts, which is a much better proxy for discretionary consumption. [22]

The VAR in Figure 12 is otherwise identical to the monthly VAR from Figure 11. [23]

The Australian economic uncertainty index increases about 10 points following a US uncertainty shock. This is broadly consistent with the findings in Section 3.2. [24]

AIC recommends two lags; BIC recommends one. [25]

Using other measures of investment, such as the RBA's estimate of non-mining private business investment (Elias and Evans 2014) or total private business investment, roughly halves the reduction in investment growth, although these are somewhat less volatile series. Other responses are essentially unchanged. [26]

As with M&E investment growth, the initial peak in the quarter following the shock is surprising in light of the sharp and more persistent reversal. [27]

Bloom (2013, 2014) provide comprehensive summaries of the empirical literature. [28]

Baker and Bloom (2013) also use an aggregate-level VAR framework, but incorporate cross-country data and use natural disasters to identify uncertainty shocks. They find much larger effects than I find, but this divergence might reflect differences between Australia and the 60 countries included in Baker and Bloom's dataset, or that the uncertainty shocks they identify – i.e. those related to natural disasters – are different to those that I identify. [29]