RDP 2024-01: Do Monetary Policy and Economic Conditions Impact Innovation? Evidence from Australian Administrative Data 4. Results
February 2024
- Download the Paper 1.23MB
4.1 Aggregate innovation measures
In this section, we examine how monetary policy shocks affect patents, trademarks and R&D activity. In particular, we focus on the coefficients of the shock variable in Equation (1) over different time horizons and trace them out in Figure 1. For consistency, all the results of this paper are based on a 100 basis point contractionary shock.
Unlike the United States (Ma 2023), monetary policy shocks have no effect on the number of patents filed (Figure 1, top panel). This is consistent with Australia tending to be an importer of new technologies, rather than a producer (Majeed and Breunig 2023) and having fewer patents filed compared to the United States. However, when focusing on somewhat broader measures of innovation there is evidence that monetary policy can have a significant influence.
Notes: Results from local projection model of aggregate innovation metrics on monetary policy shock. Trademark sample excludes 1994–1995 due to break in time series. Patents and trademarks include only those with Australian filers. Models have eight lags of growth in GDP, TWI and CPI, and four lags of shocks and dependent variable. Results robust to other specifications. Dashed lines show 90 per cent confidence intervals, with Huber–White standard errors.
Sources: ABS; Authors' calculations; IP Australia.
In the year following a 100 basis point contractionary shock, R&D spending declines by around 5 per cent (bottom panel), while the number of trademarks falls by around 15 per cent one to two years after the shock (middle panel; marginally significant). The response of R&D spending is somewhat larger than documented in Moran and Queralto (2018), though it is also shorter lived – they document an 0.8 per cent decline, with the peak effect around four years after the shock. In part, this appears to reflect our use of a local projection model instead of a VAR. When we estimate a simple five-variable VAR where the shock is ordered first, as suggested by Plagborg-Møller and Wolf (2021), the peak effect is more sustained (see Figure B1). However, the peak effect is still larger and occurs earlier compared to Moran and Queralto (2018).[7] Our results are more in line with those from Ma and Zimmerman (2023), who also use local projections and find that aggregate investment in intellectual property falls by around 1 per cent one to two years after the shock, and that listed company R&D spending falls by around 3 per cent, with a peak effect two to three years after the shock. More generally, the 5 per cent response, while large, is not unreasonable in the context of this series – over the sample the year-ended growth rate of R&D spending has ranged between around –8 per cent and 22 per cent, with a standard deviation of around 7.5 per cent.[8]
4.2 Firm-level innovation and adoption measures
As discussed above, much of the literature to date has focused on relatively narrow measures of innovation that are unlikely to fully capture the broader effects of monetary policy on adoption of existing technologies by firms. This is even though technology adoption is a crucial mechanism in models such as that of Moran and Queralto (2018).
Focusing on our broader survey measure of innovation we find that a contractionary 100 basis point monetary policy shock has relatively little effect on the overall share of firms innovating (Table 1). But this average result hides very different outcomes for small and large firms. The share of SMEs (less than 200 employees) innovating falls by 6 percentage points the year after a 100 basis point contractionary monetary policy shock (Table 1). This equates to around 52,000 fewer firms innovating (based on the number of firms with 1–200 employees in 2019/20).
| Year 0 | Year 1 | Year 2 | Year 3 | |
|---|---|---|---|---|
| All firms | ||||
| Effect | −0.76 (1.65) |
−2.76* (1.47) |
1.29 (2.09) |
7.06** (2.67) |
| R2 | 0.20 | 0.14 | 0.11 | 0.09 |
| No of observations | 45,053 | 35,635 | 26,302 | 17,536 |
| SMEs | ||||
| Effect | −2.85 (1.61) |
−6.13*** (0.99) |
−2.45 (1.83) |
1.39 (1.54) |
| R2 | 0.19 | 0.14 | 0.12 | 0.10 |
| No of observations | 29,551 | 22,185 | 14,616 | 7,291 |
| Large firms | ||||
| Effect | 3.47 (2.64) |
2.70 (2.65) |
5.02* (2.35) |
9.80* (4.46) |
| R2 | 0.19 | 0.11 | 0.07 | 0.06 |
| No of observations | 15,502 | 13,450 | 11,686 | 10,245 |
| Difference by size significant at: | 5 per cent level | 1 per cent level | 1 per cent level | 10 per cent level |
|
Notes: Significance assessed using T ‐ distribution with t – n degrees of freedom as suggested by Cameron and Miller (2015) to account for small number of clusters, where t is sample length and n is number of coefficients. All regressions include controls for industry, (lag) GDP growth, (lag) inflation, (lag) growth in the exchange rate, (lag) turnover growth and (lag) employment, and lag of the shock and dependent variable. ***, ** and * denote statistical significance at the 1, 5 and 10 per cent levels, respectively. Standard errors are shown in parentheses and are clustered at an annual level. |
||||
In contrast, the share of larger firms (more than 200 employees) innovating increases by 5 percentage points two years after the shock, equating to around 200 firms (Table 1). This latter finding may seem surprising in the context of the existing literature. But it is consistent with some of the broader innovation literature that argues that innovation may be countercyclical, as firms may have the incentive to innovate when input costs are relatively cheap and when demand is weak (Aghion et al 2012). Larger firms may be better placed to take advantage of countercyclical conditions.
To put these numbers into context, from 2005/06 to 2019/20 the share of firms innovating rose consistently, increasing by around 7 percentage points by the end of the sample (DISER 2021). Given this, monetary policy appears to have meaningful effects on the share of firms innovating.[9] Moreover, the heterogeneity of results suggests that firms may be differentially exposed to different monetary policy channels. We explore this in Section 5.
4.3 Robustness
In this section, we test to see if our results differ substantially when we use other monetary policy shock measures. We consider four measures:
- The change in the policy rate (cash rate) itself.
- A version of the Beckers (2020) shock that does not purge market expectations, which we call ‘Beckers’.
- A Romer and Romer-style shock where the policy reaction function includes only economic variables and excludes the financial market variables used in Beckers (2020). This is taken from Bishop and Tulip (2017). We refer to this as ‘BT’.
- A measure based on high-frequency changes in bond yields based on a 90-minute window around announcements (Hambur and Haque 2023), which we call ‘levels shock’.
The results are provided in Tables B1 to B4. Using the Beckers variable gives qualitatively and quantitatively similar results to those we discuss above, consistent with the high correlation between the two shock measures. The results with the BT variable and the change in the policy rate are also qualitatively similar, with SMEs decreasing their innovation and large firms increasing, though the effects are not statistically significant. Interestingly, the evidence for the levels shock is qualitatively different, with some evidence of a decline in innovation for large firms. However, given the shortcomings with this measure discussed in Hambur and Haque (2023), the results from the Romer and Romer-style shocks remain our preferred estimates.
Footnotes
We experimented with a VAR identified using timing restrictions as in Moran and Queralto (2018). This provided no evidence of a significant effect on R&D spending, likely reflecting the strong (and likely inaccurate) identification assumption. [7]
Note that the unconditional standard deviation is much larger than the standard deviation of the structural R&D shock from the small VAR discussed later, suggesting much of the volatility reflects factors such as aggregate demand. [8]
The response may seem large relative to how much the share of innovating firms in the economy varies. But it is important to keep in mind that a 100 basis point monetary policy shock is larger than any shock which has occurred in the sample. [9]