Climate change and monetary policy: a Bayesian DSGE perspective

Households, identified by z, are uniformly distributed across the interval from 0 to 1. A fraction \(1-\lambda \) of these households, known as Ricardian, optimize their consumption over time. The remaining fraction \( \lambda \), referred to as non-Ricardian households, are constrained by their liquidity and thus consume exclusively from their current income and endowments. Each household is characterized by the same utility function:where E stands for the expectation operator, \(\beta \) denotes the discount factor, \(\epsilon _{t}^{TP}\) represents time preference, \(C_{t}\) denotes a composite index of private consumption, formulated as \(C_{t}=\left[ \int _{0}^{1}C_{t}^{ }(z)^{\frac{\theta -1}{\theta }}dz\right] ^{\frac{ \theta }{\theta -1}}\), with \(C_{t}(z)\) being the consumption of good z and \(\theta \) the elasticity of substitution between different goods. \(N_{t}(z)\) indicates the labor hours supplied, and \(\varphi \) is the Frisch elasticity of labor supply. The time-preference shock is a commonly employed method to model demand-side shocks in DSGE models. The time preference follows a log-linear AR(1) process: \(\hat{\epsilon }_{t}^{TP}=\rho ^{TP}\hat{\epsilon } _{t-1}+\hat{\varepsilon }_{t}^{TP}\), where \(\rho ^{TP}\) is the persistence parameter, \(\hat{\varepsilon }_{t}^{TP}\) is a normally distributed i.i.d. error term with mean zero. The notation with a hat signifies the percentage deviation from the steady state in a log-linear model. Ricardian households, receiving dividends from firms and facing government-imposed income and consumption taxes, adhere to the following resource constraint:\(N_{R,t}\) and \(C_{R,t}\) denote the labor supply and consumption of Ricardian households, respectively. The model also incorporates \(B_{t}\) which represents the nominal price of a government bond that pays off $1 upon maturity at \(t+1\) and \(r_{t}\) the nominal yield of the bond. The nominal wage is indicated by \(w_{t}\), while \(v_{t}\) captures the financial returns from firms that fully impute dividends, with these returns being subject to taxation in a manner analogous to household income. The model specifies tax rates on income and consumption with the symbols \(\tau ^{y}\) and \(\tau ^{c},\) respectively. The return on private capital is denoted by \( r_{t}^{k}\), and the general price level is represented by \(P_{t}\), which iswhere \(P_{t}^{ }\left( z\right) \) signifies the price of individual good z. Investment by the private sector is indicated by \(I_{t}\) and it incurs quadratic adjustment costs modeled as \(\phi (\cdot )=\) \(\phi /2(I_{t}/K_{t}-\delta )^{2}\), where \(\delta \) represents the depreciation rate of private capital. The formula for the private capital stock is given by \(K_{t+1}=(1-\delta )K_{t}+I_{t}\).

The optimal behavior conditions for Ricardian households can be summarized as follows:where \(\Lambda _{t,t+1}=\beta \bigg (\frac{C_{R,t}}{C_{R,t+1}}\bigg )\) is a factor adjusting for future utility from consumption, and this equation balances the cost and returns of capital investment, inclusive of taxes and depreciation. It is assumed that the log-linearized investment equation includes investment shocks (IS) that follow an AR(1) process similar to other shocks within the model.

For non-Ricardian households, who derive their income from employment in firms and government transfers and are subject to taxation, the following conditions dictate their optimal behavior:The aggregate consumption and labor supply for the economy are then calculated by weighting the consumption and labor supply of both non-Ricardian and Ricardian households according to their respective proportions in the population, as: \(C_{t}=\lambda C_{N,t}+(1-\lambda )C_{R,t} \) and \(N_{t}=\lambda N_{N,t}+(1-\lambda )N_{R,t}\).

TFP exhibits a strong correlation with GDP. For instance, Fernald and Wang (2016) identify a correlation coefficient ranging between 0.64 and 0.74 for the relationship between TFP and output using quarterly data from the period 1984–2015. To incorporate endogenous and procyclical TFP within a DSGE model framework, one can employ a learning-by-doing equation, as demonstrated in the works of (Chang et al. 2002; Engler and Tervala 2018), and Watson and Tervala and Watson (2022):The notation \(Y_{t}(z)\) s used to represent the output produced by firm z, while \(K_{G,t}\) stands for public capital, with \(\phi _{kg}\) representing its output elasticity. The term \(A_{t}\) is used to describe TFP, capturing the combined level of workforce skill and other efficiency-enhancing factors. The evolution of TFP is attributed to a learning-by-doing mechanism, influenced by historical labor inputs, and is mathematically formulated as:where \(\rho _{x}\) captures the persistence of the TFP stock over time, and \( \mu _{l}\) quantifies the responsiveness of TFP to the labor hours worked in the previous period.

Minimizing costs results in a capital-to-labor ratio given byFirms strive to optimize the present value of their future profits, which is represented as \(v_{t}(z)\)with \(1-\gamma \) indicating the probability that a firm can revise its prices in any given period. Utilizing the stochastic discount factor \(\xi _{t,s}\) for the interval between t and s, the optimal price setting, \( p_{t}(z)\), is determined aswhereApplying log-linearization to Eq. (11) results in the pricing formula \(\hat{p}_{t}(z)=\beta \gamma E_{t}(\hat{p}_{t+1}(z))+(1-\beta \gamma )(\hat{MC}_{t}(z))+\epsilon _{t}^{CP}\), where \(\epsilon _{t}^{CP}\) denotes a cost-push shock, following an AR(1) process similar to other shocks in the model. The overall price level can be represented as \(\hat{p}_{t}=\gamma \hat{p}_{t-1}+(1-\gamma )\hat{p}_{t}(z)\).

The core theme of this study is the consequences of business cycles and monetary policy on emissions. Heutel (2012) represents the first empirical investigation into the response of CO2 to cyclical variations in GDP. He assumes that domestic emissions are a function of GDP and estimates the elasticity of emissions with respect to GDP to be 0.7. We follow this modeling approach, where the log-linearized version of the flow of emissions, \(F_{t}\), is characterized by the equation \(\hat{F}_{t}=\zeta \hat{ Y}_{t}^{ }+\epsilon _{t}^{F}\). Here, \(\zeta \) denotes the elasticity of emissions with respect to GDP, and \(\epsilon _{t}^{F}\) represents an emissions shock. It follows an AR(1) process similar to other shocks in the model. It can be interpreted as fluctuations in the level of emissions that are not attributable to the cyclical variations in GDP. These fluctuations may arise, for example, due to changes in the composition of GDP affecting emissions.

Drawing from insights in both chemistry and economics, our modeling approach for the stock of pollution follows (Heutel 2012; Khan et al. 2019). The log-linearized version of the pollution stock, denoted as \(X_{t}\), is represented bywhere \(\eta \) represents the decay rate of CO2, calibrated based on the “half-life” of atmospheric CO2. The equation diverges from Heutel (2012) and Khan et al. (2019) by exclusively accounting for the emissions of a single country, omitting emissions from the rest of the world. Consequently, the stock of CO2 pollution should be interpreted as the quantity of emissions attributable to Australia.

Environmental economics research occasionally explores the bidirectional relationship between climate change and the global economy (e.g., Nordhaus (2014)). Global economic activity generates CO2 emissions, which influence climate change, while climate change in turn affects global economic activity. However, Australia’s CO2 emissions have a minimal impact on the global CO2 stock driving climate change. This limitation prevents the development of a feedback loop model that addresses the Australian economy.

It is posited that public consumption indices parallel those of private consumption, with the demand functions for domestic goods by the public sector being structured analogously to those of the private sector. The equation describing the formation of public capital is identical to that governing the formation of private capital. The government’s budget constraint is outlined as follows:The government adjusts the income tax rate in response to deviations in the previous period’s public debt from its initial steady state level:where \(\Phi _{d}\) signifies the tax elasticity with respect to public debt. Government spending across various categories, denoted as \(\hat{g}\) (where g represents \(G^{C}\), \(I^{G}\), and \(G^{T})\) follows an AR(1) process: \( \hat{g}_{g,t}=\rho ^{g}\hat{g}_{g,t-1}+\varepsilon _{g,t}^{g}\). In this formula, \(\rho ^{g}\) varies between zero and one, \(\hat{g}_{g,t}\) is expressed as \((G_{g,t}-G_{g,SS})/Y_{SS}\), and \(\epsilon _{g,t}^{g}\) represents an independent and identically distributed spending shock with a mean of zero.

Monetary policy adheres to a (Taylor 1993) rule, which includes the aspect of interest rate smoothing. The log-linearized formulation of the monetary policy rule is given by:where \(\Delta \) signifies the first difference operator, and \(\hat{ \varepsilon }_{t}^{r}\) represents a monetary policy shock with a mean of zero.

Aligned with conventional parameters in DSGE literature, the discount factor \((\beta )\) is fixed at 0.995, and the elasticity of substitution between goods \((\theta )\) is determined to be 6. The Frisch elasticity for labor supply \((\varphi )\) is determined to be 1. The output elasticity of private capital \((\alpha )\) remains standard at 0.33. Based on Watson and Tervala (2022) estimation for Australia, the output elasticity of public capital (\( \phi _{kg}\)) is set at 0.084. The depreciation rates for public and private capital on a quarterly basis are 1.25% and 2.5%, respectively, translating to annual rates of approximately 5% and 10% in accordance with the (IMF 2015). Non-Ricardian households \((\lambda )\) are accounted for 27% of the population, mirroring (Watson and Tervala 2022), based on the proportion of Australian households without debt. The consumption tax rate \(( \tau ^{c})\) is fixed at 10%, reflecting Australia’s Goods and Services Tax rate as noted by the OECD (2020). Income tax rate (\(\tau _{t}^{y}\)) is set at 27%, reflecting the average tax wedge for a single worker during the study period (OECD 2021), consistent with (Watson and Tervala 2022).A key parameter concerning the stock of emissions is the decay rate of CO2 \(( \eta )\). Heutel (2012) utilizes a base case of 83 years for the half-life of CO2, based on Reilly (1992). It implies a quarterly parameter \(\eta \) of 0.9979. However, this parameterization deviates from the Intergovernmental Panel on Climate Change (IPCC) perspective as articulated by Solomon et al. (2007), which suggests that approximately half of a CO2 pulse to the atmosphere is eliminated over a period of 30 years; an additional 30% is eliminated within a few centuries; and the remaining 20% typically remains in the atmosphere for many thousands of years. We set the quarterly decay rate of CO2 to 0.9942, such that half of the CO2 pulse is removed over a period of 30 years (120 quarters).

Our initial data point corresponds to Australia’s National Greenhouse Gas Inventory’s (2023) dataset, which provides seasonally adjusted, weather-normalized emissions starting from 2004:Q3. Due to the exceptional period of the COVID-19 pandemic, we truncate the data at 2019:Q4. AustraliasNational (2023) data on seasonally adjusted, weather-normalized emissions is expressed in million tonnes of carbon dioxide equivalent. We utilize data on quarterly real GDP, private consumption, private investment, the consumer price level, and the interest rate. The first three datasets are sourced from the National Accounts by ABS (2023a), the price level from ABS (2023b), and the quarterly interest rate is derived from the monthly average of the overnight cash rate (RBA 2023).

All data are decomposed using the Hamilton filter (Hamilton 2018), which is well suited for this study due to its ability to avoid over-smoothing and end-point bias, issues commonly associated with the Hodrick–Prescott (HP) filter. These limitations of the HP filter can remove critical long-term information, particularly in emissions data. By relying solely on lagged values, the Hamilton filter preserves long-term dynamics and ensures trend estimates are based only on past information. This approach provides a reliable framework for extracting cyclical and trend components from seasonally adjusted, weather-normalized emissions and macroeconomic variables, while addressing concerns about time-varying trends.

According to the IMF’s Global Debt Database (Mbaye et al. 2018), the average public debt in Australia from 2004 to 2019 was 26% of annual GDP. In our model, the public debt ratio is set to 104% relative to quarterly GDP. During the period from 2004:Q3 to 2019:Q4, public consumption averaged 19% and public investment 3% of GDP. In the model, the ratio of public consumption and investment to GDP is set to match these figures. In our model, the shares of GDP allocated to private consumption and investment are determined endogenously through the model’s parameters and are 62% and 17%, respectively. According to the data (ABS 2023a), these shares are 52% and 20%, respectively, indicating that the model’s allocation for private consumption is higher than observed in the data. However, it is worth noting that due to the contribution of net exports (5% of GDP) in the data, the figures do not sum up to 100%, and due to rounding, the model’s GDP shares in these calculations total 101%.

The rest of the model parameters are estimated using Bayesian inference techniques. The stochastic behavior of the model is shaped by seven exogenous shocks: shocks to public consumption (\(\sigma ^{C^{G}})\) and investment \((\sigma ^{I^{G}})\), monetary policy \((\sigma ^{r})\), private investment \((\sigma ^{IS})\), time preference \((\sigma ^{TP})\), cost-push \(( \sigma ^{CP})\), and emissions \((\sigma ^{F})\). We interpret the time preference shock as a private demand shock.

The prior for the elasticity of emissions relative to GDP \((\zeta )\) is set at 0.7, based on Heutel (2012). The parameter follows a normal distribution with a standard deviation of 0.2.

The prior for the persistence of TFP \((\rho _{x})\) and the elasticity of TFP with respect to past employment \((\mu _{l})\) are set at 0.89 and 0.18, respectively, based on (Watson and Tervala 2022). The persistence of all AR(1) processes follows a beta distribution, with a standard deviation of 0.05 for the persistence of TFP. The elasticity of TFP and similar parameters follow a normal distribution with a standard deviation of 0.1.

Smets and Wouters (2007) established the priors for the persistence of all shocks at 0.5. However, Watson and Tervala (2022) found higher values for public expenditure shocks. We set the persistence of public consumption \(( \rho ^{C^{G}})\) and investment shocks \((\rho ^{I^{G}})\) 0.75 with a standard deviation of 0.1. The persistence of monetary policy \((\rho ^{r})\), private investment \((\rho ^{IS})\), time preference \((\rho ^{TP})\), cost-push \((\rho ^{CP})\), and emissions \((\rho ^{F})\) remains at 0.5, with a standard deviation of 0.2. The standard deviation of all shocks is set at 0.04, with a mean of 0.05.

The prior for the Calvo parameter \((\gamma )\) is established at 0.65, informed by the study conducted by Cagliarini et al. (2011), specifically addressing the value of the Calvo parameter in Australia. Its standard deviation is set at 0.05. The priors for the parameters governing monetary policy adhere to the specifications outlined in Smets and Wouters (2007). The smoothing parameter \((\mu _{s})\) is determined to be 0.75, exhibiting a standard deviation of 0.05, distributed according to a Beta distribution. Responses to inflation \((\mu _{p})\) and output \(\mu _{y})\) follow a Normal distribution, with mean values of 1.5 and 0.125 (one-quarter of 0.5), and standard deviations of 0.1 and 0.05, respectively. The prior for the investment adjustment cost parameter \((\phi )\) is fixed at 4, as in Smets and Wouters (2007), accompanied by a standard deviation of 1.

We initialize the Metropolis–Hastings (MH) algorithm chain at 0.3, with a scale parameter of 0.45 for the covariance matrix in the random walk MH algorithm, resulting in a 28% acceptance rate. This aligns closely with the optimal acceptance ratio of 23% recommended by Roberts et al. (1997) and falls within the optimal range of 25%–33% advocated by Adjemian et al. (2024). Five parallel chains are utilized for the MH algorithm, in line with the approach of Smets and Wouters (2007), and the algorithm undergoes 250,000 Markov chain Monte Carlo replications. We discarded the initial 50% of draws (125,000). Table 1 shows the posterior means and credible intervals of parameters, obtained through the 95% highest density interval (HDI). Figure 2 illustrates prior distributions of key variables in gray and posterior distributions in black.

In the context of the study’s framework, the pivotal parameter is the elasticity of emissions relative to GDP, with a posterior estimate of 0.52. This implies that a 1% cyclical change in GDP correlates with a 0.52% cyclical change in emissions. Heutel (2012) was the first to analyze the relationship between emissions and GDP cycles, utilizing seasonally adjusted monthly data from the USA and finding a value of 0.696. This value has been employed in numerous macroeconomic models in environmental economics. Khan et al. (2019), using US data, determined that the correlation of the cyclical components of GDP and emissions, based on quarterly data spanning the past 40 years, is 0.64.

Our estimation yields a lower value, potentially due to several factors: a decline in the correlation between GDP and emissions over time, as our dataset begins from a later period; differences between countries; or the use of weather-normalized emissions data. The weakening relationship between emissions and GDP over time likely reflects the impact of climate policies and technological advancements in reducing the carbon intensity of economic activity. These same factors, which have driven the long-term decoupling of emissions from GDP, may also explain the reduced correlation between emissions and cyclical fluctuations observed in our study.

The estimates support the ““learning-by-doing” perspective, suggesting that fluctuations in employment affect TFP. The posterior persistence of TFP is not particularly high and thus variations in employment are not associated with a long-lasting, hysteresis effect on TFP. Nonetheless, the learning-by-doing equation effectively captures the procyclical nature of TFP. On the other hand, the confidence interval for the elasticity of TFP with respect to past employment is relatively wide.

The posterior estimates for the persistence of fiscal shocks range from 0.4 to 0.6, while for other shocks, it ranges from 0.5 to 0.7. The posterior persistence of emission shocks is only 0.2. The posterior estimation indicates that the Calvo parameter is 0.55, implying an average price stickiness of around two quarters. Comparatively, Bayesian DSGE models calibrated with Australian data have yielded values of 0.38 (Li and Spencer 2016), 0.71 (Watson and Tervala 2022), and 0.79 (Justiniano and Preston 2010), placing our estimate at the lower end of this spectrum.

Our posterior estimate of the monetary policy smoothing parameter (0.72) is fairly typical but slightly lower than the 0.84 estimated for Australia by Langcake and Robinson (2018) and Justiniano and Preston (2010). The coefficient for output in the monetary policy rule in our model is 0.2. Justiniano and Preston (2010) and Langcake and Robinson (2018) estimate slightly different rules, incorporating both output gaps and economic growth. However, a common finding is a relatively strong reaction to output changes. For inflation, our model estimates a coefficient of 1.6, which is slightly above (Justiniano and Preston 2010) estimate of 1.46 and somewhat below (Langcake and Robinson 2018) value of 1.83.

To understand the relationship between business cycles and emissions, we begin by analyzing the economy’s and emissions’ response to a private demand shock (time-preference shock). Figure 3 displays the impulse responses of key variables, including posterior means and 95% credible intervals, to a demand shock that reduces output by 1% in the first period.

In the model, a positive demand shock triggers an uptick in consumption and labor supply, resulting in increased output. The subsequent boost in employment, driven by the learning-by-doing mechanism, amplifies the expansion phase. As the economy enters an upswing, inflation ensues, prompting monetary policy to counteract with interest rate hikes to temper inflation and restrain the boom. Additionally, the increase in employment fosters higher TFP in the medium term, further fueling the boom.

Despite the inclusion of a learning-by-doing equation in the model, which could potentially induce hysteresis effects in TFP and output, the posterior estimate for TFP persistence (0.88) suggests that changes driven by employment are relatively short-lived. The half-life of TFP is approximately five periods, meaning half of a shock’s impact dissipates within this time, and the effect declines to just 35% of its initial magnitude within eight periods. This indicates that the effects of shocks on TFP diminish rapidly after the initial period, implying that the medium-term expansion in output is not driven by TFP hysteresis.

In the model, the fiscal balance is endogenously determined, fluctuating in tandem with the business cycle as tax revenues adjust to output fluctuations. An increase in tax revenues leads to a reduction in public debt, triggering a decline in the tax rate as per the tax rule. This fosters employment and encourages investment. The subsequent rise in employment boosts TFP in the medium term, resulting in a sustained increase in output.

A key focus of the study lies in examining the connection between business cycles and emissions. Initially, there is a 1% decrease in output, accompanied by a corresponding 0.52% rise in emissions. This is due to the estimated elasticity of emissions relative to GDP being 0.52. By the 20th period, there is a 0.19% increase in output and a 0.1% increase in emissions, indicating sustained growth in both GDP and emissions within the model. It should be noted that the change in GDP over the medium term is statistically significant.

A private demand shock has a significant impact on the CO2 stock. Initially, it leads to a persistent rise in emissions. Additionally, due to the extended lifespan of CO2, emissions continue to affect the CO2 stock for an extended period. By the 20th period, the CO2 stock has increased by 2.6%. It is worth noting that this model is closed-economy, so the CO2 stock should be understood as the accumulation resulting from the Australian economy.

The monetary policy shock, whose effects on the main variables are depicted in Fig. 4, represents a 100 basis point increase in the annualized interest rate in the first period. It decreases output by 0.84%. Lane (2023) provides a summary of the ECB’s models, which estimate that a 100 basis point change in interest rates leads to an average output effect of around 1%. However, the range of estimates (approximately 0.25–1.25%) is large, depending on the methodology used.

In the model, monetary policy leads to traditional effects such as a decline in private consumption and investment. However, two distinctive features of the model concerning monetary policy are endogenous TFP and fiscal balance. A decrease in employment initially triggers a decline in TFP due to the learning-by-doing equation. This aligns with (Baqaee et al. 2024), who empirically demonstrate that monetary shocks significantly impact TFP, generating procyclical, hump-shaped movements and further supporting the view that TFP is endogenous and sensitive to demand shocks.

An increase in interest rates raises government interest payments. Additionally, the recession caused by the interest rate hike reduces tax revenues, leading to an increase in government debt. The model incorporates a tax rule that raises income taxes, thereby reducing investments and employment, consequently lowering TFP in the medium term. As TFP, private capital, and employment decrease in the medium term, output also declines. However, it should be noted that the impact on GDP is relatively small at 0.12%.

From Figure 4, it can be observed that emissions decrease by 0.43% in the short term and by 0.09% in the 20th period. The decline in emissions is a result of the reduction in output. As previously noted, the estimated elasticity of emissions relative to GDP is approximately 0.5, indicating that emissions change about half as much as GDP. The monetary policy shock induces a long-lasting effect on the pollution stock, as the production impact is enduring and CO2 has a very long lifespan. The stock of pollution decreases by 2.1%. However, it is noteworthy that the confidence interval for this estimate is quite large.

Attílio et al. (2023) used the GVAR methodology to estimate the CO2 effects of monetary policy in the USA, UK, Japan, and the Eurozone. They find that monetary contractions typically reduce emissions, with a 0.4% short-term reduction in the USA and a statistically insignificant 0.2% long-term effect. Our study finds that a 100 basis point increase in the annualized interest rate in the first period, much smaller than a one-standard-deviation monetary policy shock, leads to a 0.43% decrease in emissions in the short term and 0.09% in the 20th period.

Qingquan et al. (2020) used panel techniques to assess the impact of monetary policy on CO2 emissions in Asian economies, finding that a 1 percentage point change in the real interest rate affects emissions by 0.11–0.19% in the long term. Our estimation suggests a larger short-term effect (0.43%), but a smaller medium-term impact (0.09%) compared to their estimate. Chishti et al. (2021) found that a 1 percentage point change in interest rates affects CO2 emissions by 0.35% in the long term in BRICS economies, using various panel estimators. However, this result seems implausibly large, as with an elasticity of emissions relative to GDP of 0.5, a monetary policy shock should affect output by 0.7%, not 0.35%, over the long term, which appears unlikely..

Bayesian DSGE models have been used to analyze shocks’ historical contributions to GDP and inflation. Our study extends this approach to examine cyclical variations in emissions. However, it should be noted that in the model, emissions vary due to both GDP fluctuations and emission shocks, with GDP fluctuations driven by factors other than emission shocks. Figure 5 depicts the quarterly historical decomposition of emissions. The black line illustrates the deviations of emissions from the trend, while the bars of various colors indicate the contributions of shocks to emission fluctuations.

Figure 5 shows that cyclical fluctuations in emissions in Australia are primarily driven by the effects of monetary policy shocks, private demand shocks, emission shocks, and public consumption and investment shocks. These factors account for the majority of the historical decomposition of emissions. In contrast, the contribution of cost-push shocks and private investment shocks is minimal, and they do not play a significant role in the cyclical fluctuations of emissions. The significant role of demand shocks in Australian business cycle fluctuations is consistent with (Rees et al. 2016). In an open economy model, he finds that domestic demand shocks account for a large portion of the variance of output growth in the non-tradeable sector.

Before the global financial crisis, positive private demand shocks drove emissions higher, while contractionary monetary policy tempered their growth. At the onset of the crisis, private demand shocks turned sharply negative, leading to a significant decline in emissions. Monetary policy became expansionary, but its impact on emissions remained muted due to the collapse in private demand.

The global financial crisis had a relatively mild impact on Australia compared to many other countries. Shortly after the deepest phase of the crisis, the recovery in private demand and the increase in public consumption typically contributed to higher emissions. In contrast, monetary policy and public investment played a moderating role, counteracting some of the upward pressure on emissions.

After the global financial crisis, private demand shocks have been the important driver of emissions variability, initially negative but turning positive as demand recovered, leading to higher emissions. Monetary policy shocks have moved countercyclically to private demand shocks, moderating emissions fluctuations. From 2015 to 2017, the relative share of emission shocks increased slightly, suggesting that structural factors, such as regulations or energy shifts, contributed to emissions variability.

Rees et al. (2016) found that contractionary monetary policy before the global financial crisis significantly reduced output. Moreover, expansionary monetary policy provided substantial support to the economy during the global financial crisis and turned contractionary in the early 2010s as the Australian economy recovered strongly. While our findings partly support this view, it is important to note that Rees et al. (2016) does not sufficiently account for the role of fiscal policy. Our results suggest that fiscal policy plays a much larger role in business cycle fluctuations in Australia than previously indicated in the literature.

Figure 5 shows that while monetary policy shocks account for a substantial share of cyclical emissions fluctuations, they are not the sole driving factor. However, the interaction between monetary policy and private demand is the dominant cyclical mechanism influencing emissions. The figure suggests a moderate negative correlation between private demand shocks and monetary policy shocks, reflecting countercyclical policy responses. Periods of strong private demand, which tend to increase emissions, have often coincided with contractionary monetary policy, mitigating these effects. Conversely, expansionary monetary policy has been implemented during downturns, supporting emissions. However, monetary policy has at times only partially offset demand-driven emissions fluctuations, allowing other economic forces to shape emissions dynamics.