2.1 Central bank models

The natural rate of interest is a determining factor in CBMs. A widely cited CBM for r* is the semi-structural model of Laubach & Williams (2003) (LW, 2003; see Holston et al., 2017 (HLW, 2017) for a novel specification and readily available quarterly estimates). They specify r* as ${\mathit{r}}_{\mathit{t}}^{*}={\mathit{g}}_{\mathit{t}}+{\mathit{z}}_{\mathit{t}}$. It assumes that r* is a function of two random walk processes, the trend growth rate of potential output, g, and a time-varying unobserved component, z. Variable g is the expected growth rate of potential output, of which TFP growth is a main driver. Thereby it reflects expected income growth of economic agents, like in the Ramsey model. Variable z captures the effects of the other factors in the Ramsey model: that is, the time preference and substitution (or smoothing) parameters. Hence, variable z includes behavioral factors that determine borrowing and saving by economic agents. Through these financial channels, the changes in agents' preferences can influence r*. Figure 1 presents these relationships in a stylized framework of a CBM.

In CBMs, r* is also a determinant of aggregate demand via the investment-saving (IS) or demand curve according to the New Keynesian framework on which CBMs are based. The channel for this demand effect is the difference between the policy interest rate and the natural rate (i.e., the rate gap). If r* is high (low) the monetary policymaker has much (little) room to stimulate the economy in a downturn by widening the real rate gap. This aspect has become relevant since the global financial crisis of 2008 (GFC), when the decline of r* and the increased likelihood of hitting the effective lower bound of the policy rate reduced the policy space for central banks to provide monetary accommodation. Through the rate gap channel, monetary policy determines demand, including investment. So, r* directly affects monetary policy (arrow at the top of Figure 1, pointing from r* to i). Besides, monetary policy can have an impact on r* as well (left-hand side of Figure 1). This refers to the effect of monetary policy on productivity growth, via hysteresis effects and the influence on the allocation of capital. In Section 12, we explain how such channels make r* endogenous to monetary policy.

Since r* is not directly observable, central banks rely on several types of model estimates of r*. One of the most widely used is the LW (2003) model. It is based on a semi-structural model that filters r* out of the (short-term) market interest rate and potential output estimates. The research by LW (since extended by HLW [2017]) shows that estimates of the natural rate of interest are highly inaccurate and can vary widely depending on the model specification. Another strand of CBMs is time series models, which proxy r* as the long-term trend in real interest rates. This trend is filtered out of the data, as in Johannsen & Mertens (2018). Other CBMs are general equilibrium models, in which r* is the interest rate at which the economy is in equilibrium while prices are fully flexible (e.g., Del Negro et al., 2017). The natural rate of interest is not necessarily constant in this equilibrium but can fluctuate due to shocks in aggregate demand and productivity or due to changes in preferences. Instead, shocks to the risk premia are usually assumed to be exogenous, which limits the explanatory power of general equilibrium models for r*.

Another limitation is that CBMs are backward looking and do not consider the adverse impacts stemming from climate change (see for instance Brand et al., 2018, 2021): however, both g and z could potentially be impacted by it. In part, this has to do with a horizon difference in CBMs and CEMs. For example, the “long-term” for central banks is shorter than the “long-term” for climate scientists (who usually calculate in time units of 5 years). Moreover, CBMs have limitations that make them inappropriate for simulating the macro-financial effects of climate risks in a comprehensive manner, in particular, the absence or limited representation of the energy sector. Recently, central banks and financial institutions have started to develop tools to better understand the macroeconomic effects of climate change (see ECB, 2021b for an overview).

2.2 Climate economy models

The class of CEMs that are based on the Ramsey growth model, such as the seminal Dynamic Integrated Climate Economy (DICE) model by Nordhaus & Sztorc (2013), consider the effects of climate damage on the discount rate, which provides a proxy for r*. The obvious channel for this relationship runs via potential economic growth g, as visualized in Figure 2. In CEMs, climate change has a negative impact on g through physical damages in a damage function which negatively affects potential output (bottom left in Figure 2). The costs of the transition (abatement costs a^C) to a carbon-neutral economy also lower potential output, by reducing disposable income at least in the initial stages and so reducing per capital consumption growth (arrows pointing up in Figure 2). Moreover, damage and transition costs lower the return on capital, which causes a decline in investment demand and reduces output. Because global temperatures increase gradually, the dampening effect on per capita consumption growth is also gradual. Abatement also has an upward effect on potential output by the additional investments needed for abatement (a^I), if he marginal product of capital exceeds the marginal cost of capital (this channel is not explicitly modeled in DICE).

In CEMs (taking DICE as an example), potential output follows from the production function in which capital and hence investment are key drivers. Investment is determined by the fraction of savings (S) from output, where S is a fixed parameter that does not depend on the interest rate. Hence, investment demand does not depend on monetary policy, such as in CBMs, implying that savings determine investment like in the loanable funds approach (see Bofinger & Ries, 2017). The only channel through which the interest rate affects the economy is the discount rate, used to determine the utility derived from labor and the social cost of carbon (SCC), as explained below. Different from CEMs, in CBMs the financial drivers of r* (captured by parameter z in the HLW version of the Ramsey equation) are not exogenous, since z is driven by a time-varying stochastic process. This reflects time-varying savings—which can be driven by financial shocks amongst other factors—which differs from the fixed savings ratio in CEMs.

Feedback effects play a significant role in the standard CEM (lhs of Figure 2). Higher output produces more greenhouse gas emissions, which feeds back negatively on the economy. This effect runs through the impact of rising cumulated emissions on temperatures, which in turn is the main input of the damage function. Over time, this effect decreases during the transition to a carbon-neutral economy. It indicates the intertemporal trade-off between the costs of improved carbon pricing in the short run and limiting physical damage in the longer run. While DICE has a limited energy sector, especially when compared to nongeneral equilibrium CEMs such as REMIND, MESSAGE, or GCAM, it is a useful benchmark for CBMs, since it covers both physical risk (via the damage function) and transition risk (via abatement costs and carbon taxes), while being based on the Ramsey growth model. Thereby DICE and similar style CEMs capture the main economic channels through which climate change and transition policies can affect r*, while they do not include the financial drivers of r* as captured by parameter z in the natural rate equation in CBMs. In that sense, CEMs and CBMs are complementary tools to assess the impact of various long-term trends on the natural interest rate.

Parameter ρ (pure rate of time preference) also plays a significant role in CEMs. In an intergenerational set-up, it reflects how agents assess the value of current consumption relative to the consumption of future generations. In CEMs, parameter ρ is used as the discount rate that determines the utility derived from labor and the SCC, that is, the current discounted value of future damages (in terms of changes in wealth) of emitting one ton of carbon today, scaled by the marginal welfare effect of an extra unit of consumption. The lower ρ, the higher the SCC, and in that sense ρ is also called an “ethical discount rate.” Bauer & Rudebusch (2020) find that the decline of the discount rate has substantially boosted the estimated economic loss from climate change in terms of the SCC. Stern (2007) argues normatively that future damages should not be discounted at all (i.e., ρ = 0). This shows that the discount rate is a key parameter in the climate-economy debate, since it determines the value of taking action to mitigate the future impact of climate change. There are also reasons for considering other issues such as inequality, risk, and human values when choosing discount rates (Fleurbaey et al., 2019). Ord (2020) criticizes the use of the discount rate in climate-economics, arguing that discounting welfare over long time horizons is an incorrect approach as it conflates monetary value with human well-being.

Other critics of current climate economy models include Weitzman (2009), who emphasizes their fundamental uncertainty, and Bolton et al. (2020), who point at unpredictable interactions and nonlinear effects that could trigger tipping points, and following from that, damages might be severely underestimated. Keen et al. (2022) show that flawed modeling assumptions can lead to a substantial underestimation of damages. Tipping points and nonlinear effects are usually not considered in climate-economy models, which tend to be deterministic. The DICE model has particularly been criticized on those grounds. Most IAMs, like DICE, do not consider economic and climate risk and instead focus on parameter uncertainty. Pindyck (2013) stressed the fact that the climate sensitivity of IAMs is surrounded by a high degree of uncertainty and that assumptions about the discount rate and risk aversion can have very decisive effects on projected outcomes. Also, the DICE model has been criticized for generating implausibly low effects of rather high temperatures, amongst others by Dietz and Stern (2015). Uncertainty about future economic and climate conditions affects the choice of policies for managing interactions between the climate and the economy. Cai and Lontzek (2022) develop a framework of dynamic stochastic integration of climate and economy (DSICE), showing that the social cost of carbon is affected by both economic and climate risks and is a stochastic process with significant variation. Therefore, the uncertainties and inaccuracies in modeling future climate damage, as well as the simplification of the representation of the climate system and the economy, call for caution about interpreting the outcomes of CEMs. This holds in particular for the impact of scenarios on climate change in the long-run. The shortcomings of CEMs are less objectionable for our purpose, which is to assess the channels through which climate change and transition policies can affect r* and thereby the economy and financial system. CEMs provide a useful frame of mind for this more analytical assessment, which does not rely on the quantification of climate scenario outcomes.

Moreover, it is important to focus on the tails of the outcome distribution as the uncertainty around the expectations makes the occurrence of tail risks more likely. Krogstrup & Oman (2019) emphasize that climate change raises uncertainty to new levels, because the tail risks include catastrophic and often irreversible events like the thawing of permafrost which could release huge amounts of GHGs and thereby intensify climate change even further. Therefore, they suggest a risk management approach inspired by Value-at-Risk (VaR) models which means maximizing one's objective function subject to the risk of catastrophic irreversible climate change remaining below a before agreed level.

3.1 Demographic trends

In recent decades, most advanced economies have experienced a demographic transition, reflecting low fertility rates, rising life expectancy, and thus changing composition of age cohorts. In the Ramsey-style CEMs, population growth influences the discount rate via per capita consumption. This is an outcome of the production function in those models, in which population and labor are main determinants. Brand et al. (2018) find that the net effect of these trends has reduced real interest rates in the euro area by around 1 percentage point since the 1980s. When extrapolating demographic trends, it is foreseen that real interest rates can be expected to fall by a further 0.25–0.5 percentage points by 2030.

For the US, estimates of the effect of demographics on interest rates over the 1970–2015 period range from a decline of less than 1 percentage point (Gagnon et al., 2021) to a more severe decline exceeding 3 percentage points (Eggertsson et al., 2019). Hence, there is substantial range estimating the magnitude of the demographic driver. On one side, Auclert et al. (2021) foresee that aging will be accompanied by higher wealth-to-GDP ratios and capital deepening, lower asset returns, putting further downward pressure on real interest rates, and widen global imbalances. Such effects will be large and heterogeneous across countries. Hence, according to Auclert et al. (2021) demographics will continue contributing to the trend decline in r* (a prediction, also shared by Gagnon et al., 2021).

On the other hand, some studies predict that future demographic developments might contribute to lowering saving rates while lifting interest rates. One is the hypothesis of a “great demographic reversal” by Goodhart & Pradhan (2020). A worsening dependency ratio, with a sharply increasing proportion of elderly and retired people, will lower savings ratios of the whole economy and hence bring about rising interest rates. This follows from the assumption of age-specific savings rates that decline in old ages as retirement thresholds are crossed (Auerbach & Kotlikoff, 1990; Bosworth et al., 1991; Summers & Carroll, 1987). Thus, demographics and an increasing share of elderly push down aggregate savings, which in turn lifts interest rates. However, climate change may reverse this trend, since extreme temperatures can have important effects on mortality, health, and, in turn, on labor supply (Seppanen et al., 2006). It could reverse the current trend towards higher life expectancy and alter the age composition of the population in favor of younger cohorts.

There is a growing literature on the effects of climate change on migration, which was surveyed by Cattenao et al. (2019). They find that climate change and its negative impact on income variability does not automatically lead to more migration, but rather climate change often acts as an effective constraint to migration. Acute manifestations of physical risks like hurricanes or floods tend to result in temporary displacements and the affected households and firms would often return to their original venues. Chronic physical risks like droughts or rising sea levels instead cause more permanent migratory outflows, often because original locations cannot be rebuilt. However, rising temperatures can also reduce international migration from poorer countries, since liquidity constraints could become more binding. While Cattenao et al. (2019) conclude that alarmist projections of future migration levels triggered by climate change are unrealistic, other scenarios of large climate-related labor supply shocks that could drive r* are also possible. Benveniste et al. (2022) quantify the effects of climate change on migration dynamics at the income quintile level and find resource-constrained international immobility for lower-income households to be the most likely outcome.

ECB (2021b) presents some arguments about how climate change could trigger migration flows and thereby affect labor supply. In the case of an outflow, an adverse impact on labor supply leads to a higher capital/labor ratio which reduces the marginal product of capital in the steady state and thus r*. The impact of climate change could differ across countries, as labor supply might rise in some countries due to immigration flows. These effects are uncertain at best and could manifest themselves over very long horizons. Summing up the overall finding in the literature is that the net impact of climate change on demographic trends is prima facie ambiguous: while it might discourage labor supply, thus exacerbating the downward effect on r*. Some studies conclude that it could also reverse the current trend toward higher life expectancy and its potential upward effect on r*.

3.2 Changes in economic growth and productivity

The literature suggests a close relationship between climate change and GDP losses, captured by trend growth variable g in the Ramsey model. The economic impact can be caused both by physical (damage) and transition risks. With regard to physical risks, a literature review by Howard and Sterner (2017) shows that the economic damage of 0.7–4.3°C temperature increases until 2100—compared to pre-industrial levels—can range from +0.1% to −23% of GDP. At the upper bound, this worst-case impact is comparable to the −25% cumulative GDP impact estimated by the NGFS (2021). Alestra et al. (2020) simulate the impact of GDP damage due to climate change and the GDP impact of mitigating measures for 30 countries, by taking a long-term supply-side view. They find that at the 2100 horizon, the global GDP loss ranges from 7% to 12%, depending on the scenario. The wide range of the simulation outcomes in the literature also reflects the uncertainties inherent to CEMs, as discussed in Section 3.

The literature distinguishes acute physical risks (e.g., extreme weather events) from chronic physical risks (due to deterioration of the environment). The latter can, amongst others, reduce labor productivity (Day et al., 2019), lower agricultural yields (Roberts & Schlenker, 2009), and abate industrial output (Cashin et al., 2014). Existing evidence finds that productivity levels can decrease in the short-term by 2% per degree above comfort temperature (Heal & Park, 2016). Day et al. (2019) review various studies analyzing the impact of global warming on labor productivity. All studies find substantial reductions in productivity for temperature increases above certain thresholds. Sectoral shocks may also vary in intensity, for example, agricultural productivity might be affected by climate change via changes in crop yields.

Both acute and chronic physical risks will affect the capital stock directly, if productive assets are damaged, or become useless, more often than under current climate conditions. This is comparable to an accelerated obsolescence rate of capital. It leads to a shift in resources from innovative activities towards reconstruction, adaptation, and prevention of further damages as described by Batten et al. (2020) and Kahn et al. (2019). Dietz & Stern (2015) suggest that damages to the capital stock will also diminish knowledge production and knowledge spillovers hence lowering the long-term growth rate of the economy. Ojeda-Joya (2022) simulates the economic effects of global warming with a state-space semi-structural model where lower productivity growth also dampens the natural interest rate. Using the flexibility of the DICE model, Ciesielski (2019) endogenized the investment decisions between capital stock and knowledge production to assess the long-term growth effects of climate change.

There can be interactions between the effects of physical risks (chronic and acute) and transition risks. With unabated climate change (e.g., in a “too little, too late” disorderly as well as in a delayed transition scenario as presented in NGFS, 2021), physical risks are also increasing over time. The delayed introduction of climate policies to mitigate and stop climate change might lead to sudden stops in carbon-intensive production and turn erstwhile productive investments into stranded assets. In this extreme scenario, both physical and transition risks can lead to increasing rates of capital depreciation, thereby reducing the capital stock and its productivity. Conversely in an “orderly” scenario in which emissions-reduction starts promptly with adequate timely climate policies, there might be a gradual divestment from carbon-intensive activities accompanied possibly by higher R&D and rapid innovation (the “Porter Hypothesis”).

With regard to transition risks, the NGFS (2021) scenarios take into account a cumulative economic impact of transition risk until the year 2100 that ranges from a positive impact (net zero 2050 scenario) to around minus 2%–3% GDP (delayed transition scenario). Either way, climate policies will change the relative price of energy inputs, which will affect the type of technologies that are developed (Acemoglu et al., 2012). Pisani-Ferry (2021) also points at an intertemporal trade-off, by underlining that green innovations and investments may improve potential output in the long run but divert resources from expenditures that drive economic growth in the short-run. While the transition requires more investment in climate-neutral innovations, these innovations might not automatically raise productivity. Instead, for additional investment to have an upward effect on productivity growth and hence on r*, the marginal product of capital should be higher than the marginal cost of capital. In fact, productivity gains of technological innovations are inherently uncertain, because they have not been tested and proven before. The transition towards a climate-neutral economy will require innovations in energy generation, mobility, and other core processes of the production sector. In the gradual process of establishing new standards of de-carbonized production, some initial investments in innovative technologies will be lost, as competing alternatives will prevail, but the clearer the new standards will emerge, the more investors' uncertainty will fall.

Some firms might not be able to afford the investments for adopting innovative technologies and will be forced out of the market. This could affect productivity either way: negatively as it might diminish the productive capital stock in the economy and thereby lower the ratio of output to invested capital or positively as those firms which are able to succeed might be more productive than their less fortunate competitors. OECD (2006) also emphasizes the possibility that climate-related regulations might act as entry barriers in certain markets and thereby reduce firm dynamics.

The issue of uncertainty is an important topic in the climate-economy literature. While climate models have improved significantly, there is still ambiguity about the likely impact of further climate change and whether future damages will be modest, catastrophic, or somewhere in between. The wide range of estimated macroeconomic effects also relates to model uncertainty, as Mercure et al. (2019) show. They distinguish models based on “equilibrium” and “non-equilibrium” theories of innovation. The former are based on neoclassical economics with optimizing agents in an economy driven by the ability to produce (supply-led). The latter assume that the economy is driven by demand (demand-led) and is in perpetual dynamic change. The two modeling strands tend to yield opposite outcomes about the economic impact of low-carbon policies. This particularly relates to the modeling of monetary and finance dimensions, and their interactions with investment, innovation, and technological change. Ramsey-style CEMs are supply-led models and thus not appropriate to simulate such interactions.

A growing literature on degrowth, post-growth, and post-development suggests that the transition to a sustainable economy does not square with more economic growth, especially in industrialized countries (Asara et al., 2015; Escobar, 2015; Kallis, 2019; Latouche, 2018). This is amongst others based on the feedback effect identified in Section 3: that is, economic growth leads to further GHG emissions, while measures designed to lower emissions may reduce economic growth. The IPCC (2022) connects climate mitigation options to multiple dimensions of well-being, beyond the traditional GDP metric, which is at the core of IAMs. The report shows that many demand side-mitigation solutions generate positive and negative impacts on wider dimensions of human well-being which are not always quantifiable.

Summing up, the overall conclusion from the literature is that the effect of climate change on total factor productivity, and hence r*, is ambiguous. Physical risks will accelerate the write-down of capital stock, especially in disorderly scenarios, and lead to less productive investments in abatement efforts to reduce damages. In addition, the increase in global temperatures will likely have a negative impact on labor productivity. Adaptation policies to soften these effects are feasible, they will likely trigger additional investment demand, thereby raising r*. Similarly, rising R&D and innovations as part of the energy transition, create demand for investment and drive-up interest rates. Climate policy will be a main determinant of these trends. Nevertheless, the transition is likely to increase the uncertainty about productivity trends, and in the literature, there is no consensus about the productivity gains of green technological innovations. Moreover, the nonlinear relation between climate change-induced physical risks and the economy implies that the uncertainties about r* will compound over time. The uncertainty is right-skewed, with negative outcomes being more likely than positive outcomes (Tol, 2018). Tipping points, which are associated with temperature increases beyond the 3–4°C range, can exacerbate the economic impact exponentially (Howard & Sterner, 2017).

3.3 Risk aversion and savings

Another channel through which climate change affects the natural rate is by possibly increasing risk aversion, working through the impact of preferences on the elasticity of intertemporal substitution. In the Ramsey model, risk aversion is captured by the smoothing preference of economic agents (parameter γ) which determines the sensitivity of r* to g. Hence risk aversion is akin to the aversion to income and consumption shocks. These shocks affect future income prospects (through variable g), inducing agents to smooth consumption across time and to save more today. Higher savings will reduce r* through an increase of the supply of savings. The study by Rachel & Smith (2017) indicates that a preference shift towards higher saving by emerging market governments, following the Asian crisis, increased desired savings which contributed to a lower natural interest rate. The flipside of higher savings is lower investments, which is also associated with economic shocks and related risk aversion. Higher risk aversion will translate into a higher risk premium which weighs on investment demand. Rachel & Smith (2017) estimate that the risk premium channel contributed to lower r*. Finance-related impacts on r* are captured by the parameter z in the Laubach and Williams model. It needs to be kept in mind that this model is a closed economy model, which does not explicitly capture the effect of international spill-overs on r*. To compensate this, the IMF devoted a special chapter in its World Economic Outlook (IMF, 2023) to the analysis of this channel, based on a structural model with an international component. The analysis finds that spillovers from the integration of global capital markets may have been powerful drivers of the natural rate, since fast growing emerging economies attracted capital from rich countries, causing capital outflows which eventually drive up r*.

Climate change may also raise uncertainties and lead to an increase in the risk premium. Climate change, adaptation to it, and the transition to a carbon-neutral economy may go in tandem with higher economic tail risk (Batten et al., 2020). Such risks are associated with fundamental uncertainty, as it depends on physical, social, and economic systems that involve complex interactions, nonlinear dynamics, and chain reactions (Bolton et al., 2020). Both physical and transition risks are inherently fat-tailed events, not reflected in past data, with a potentially unlimited downside exposure (Weitzman, 2009). Hence, climate change will be a source of more frequent, intense, and persistent shocks in the economy that are difficult to disentangle. The related fundamental uncertainty will lead to a reduced willingness to invest and a greater propensity to save. These financial channels, captured by variable z in the Laubach and Williams model, can lower r*. These endogenous financial channels are typically not part of CEMs and thus CBMs are a useful complementary tool to assess the effects of these channels on r*.

Time-varying risk aversion has been documented by several studies. The major drivers behind these variations in the risk attitude of economic agents are changes in the macroeconomic environment and not necessarily idiosyncratic negative shocks experienced by individuals. Dohmen et al. (2016) and Guiso et al. (2018) present evidence for increasing risk aversion in the aftermath of the 2008 global financial crisis. Cameron & Shah (2015) provide experimental evidence from Indonesia that individuals who have experienced natural disasters exhibit a higher degree of risk aversion because they adapt their expectations about the occurrence probability of such events. This perceived increase in further natural disasters reduces their willingness to accept additional separate risks. Sakha (2019) reports the results from repeated risk experiments in Thailand, where negative agricultural shocks like droughts or floods increased the risk aversion of households. She also finds that a negative assessment of their well-being increases individuals' risk aversion.

Another contributing factor could be higher demand for safe assets, reinforced by climate change-related uncertainty. Analyzing the impact of rare disasters (i.e., wars in the 20th century) on asset markets, Barro (2006) finds that as more frequent disasters magnify uncertainty about the future, households demand more and safer assets. For instance, an increase of uncertainty about asset valuations in the transition to a carbon-neutral economy may increase the demand for safe assets. As a result, the premium that market participants are willing to pay for holding safe assets is likely to rise, putting downward pressure on r*. This refers to the safe asset channel, which asserts that safe assets hold a convenience yield, or scarcity premium, that lowers the bond yield. The scarcity premium on safe and liquid assets—that is, highly rated government bonds—increased, as the demand for such assets outpaced supply. Several studies find that the ensuing scarcity has driven up the price of safe bonds and thereby lowered risk-free interest rates (Bernanke, 2005; Caballero & Farhi, 2014; Del Negro et al., 2017; Krishnamurthy & Vissing-Jorgensen, 2012).

An important consideration is that climate change might have asymmetric effects on risk aversion. Physical risks might affect risk aversion of agents in vulnerable regions and sectors, whereas transition risk can affect risk aversion across a wide spectrum, depending on the scenario of policy adjustments. A scenario of an erratic and uncertain adjustment path will likely elevate risk aversion and market volatility more than a clear and well-communicated climate policy path. Decarbonization policies could also lead to (temporary) disruptions in the supply of energy and commodities; for instance, if mining of certain commodities would be curtailed by stricter regulation. This could lead to increased volatility of commodity prices and raise uncertainty about investment returns and the decarbonization path.

Empirical research shows that investors translate climate-related uncertainty into a higher risk premium. The impact of risk premiums on r* can be simulated by CBMs, which capture those effects via variable z. CEMs are not appropriate for this, since risk premium which determines that the return on risky capital as compensation for nondiversifiable risk, is an exogenous parameter in these models which is thus not affected by climate-related risks. Based on a combined asset pricing in an endogenous growth model, Hambel et al. (2020) find that the risk-free interest rate decreases due to additional precautionary savings needed to cope with global warming. Fernando et al. (2021) associate climate shocks with higher equity risk premia, which according to their simulations could lead to a loss in economic activity each year of up to 2% of GDP in the worst-case scenario. Bolton & Kacperczyk (2020) find that investors in US stock markets also charge a premium for the risk of holding the stocks of disproportionately high carbon emitters. Cevik & Jalles (2020) find that climate change also affects sovereign bond yields. Countries that are more resilient to climate change benefit from lower bond yields and credit spreads relative to countries with greater vulnerability to climate change. This suggests that climate change can increase the divergence between safe and nonsafe sovereign bonds. Battiston & Monasterolo (2020) investigate the impact of climate policy scenarios on the valuation of corporate and sovereign bonds, considering endogeneity and deep uncertainty. They find that climate policy scenarios can have a substantial impact on bond yields, which would imply that also r* would be affected.

Summing up, the literature shows that the qualitative and quantitative impact on r* through financial channels is hard to quantify. Empirical research about the impact of climate change on bond yields is scarce. Nonetheless, the limited available research indicates that financial risks of climate change can put further downward pressure on r* through higher risk premiums. Given that financial markets are forward looking, this effect through z likely has a more immediate impact on r* than changing preferences of economic agents related to changing income prospects (reflected in parameter g).

3.4 Fiscal policy

Governments have the primary responsibility and the appropriate tools to spur the transition to a carbon-neutral economy and to counter the economic consequences of climate change. If the negative externalities of climate change are insufficiently priced, this market failure can be addressed by taxation and regulatory measures by governments. As the IMF (2023) shows, the impact of carbon taxes on r* depends on the design of the tax and the level of international participation in their implementation. The IMF simulates a budget-neutral implementation of a global carbon tax, using the tax revenues for transfers and public investment. Due to frictions in the model, the carbon tax depresses investment and r* more than the subsequent investment demand could raise it.

Climate change will also lead to higher government spending due to three main climate-related outlays: that is, the costs of addressing physical damages stemming from extreme climate events (such as climate-related disasters), rising adaptation costs to reduce the impacts from rising physical risks, and escalating mitigation investments to decarbonize. Moreover, as part of adaptation, government spending might also have to contribute to nondirectly productive infrastructures and grids essential for the green transition (e.g., smart-grids and digitalization efforts). Higher public spending could also be related to social security expenditure to cover health, emergency housing, relief efforts, and other costs stemming from natural disasters. The commensurate increase of fiscal deficits will likely lead to an increase of government debt and the associated higher demand for savings will exert an upward pressure on r*. The extent to which this occurs depends on the potential crowding out of private investment by higher public debt (Eggertsson et al., 2019; Rachel & Summers, 2019). Mian et al. (2021b) argue that large debt burdens can even lower aggregate demand, as a debt-financed short-run boom comes at the expense of future demand. This debt burden on demand lowers the natural rate of interest.

The impact on fiscal sustainability will depend on the timely adoption of concerted climate policies, thus on the thrust of transition scenario. An orderly sustainable transition could support TFP growth by utilizing adequate carbon pricing (through a mix of carbon taxes and tariffs, emissions trading systems [ETS], and carbon permits) with the proceeds being reallocated to support, for example, sustainable infrastructures and innovative projects and R&D. More transition measures would even contribute to higher TFP growth and so have an upward effect on r*. For instance, in the NGFS (2021) simulations, the carbon price increase in the transition scenario has an offsetting growth effect owing to carbon revenue recycling. Hence, the quality and composition of fiscal policies in the context of climate policies matters.

Through the safe asset channel, a higher supply of sovereign bonds, related to higher public debt levels, would increase r*. A higher bond supply reduces the scarcity premium, which keeps r* low in a situation with excess savings and a shortage of safe assets in which to invest. In such case, the safe asset channel may also have an upward effect on r*. In so far the scarcity premium is captured by variable z in the HLW equation for r*, CBMs provide a useful framework to assess to impact of fiscal policy on r*. CEMs are appropriate for this with regard to the effect of fiscal spending on investment and TFP growth.

Summing up, while the literature is not completely unambiguous about the impact of fiscal policy on r*, most evidence points at un upward effect stemming from an increased supply of government bonds and increasing fiscal spending that raises TFP growth.

3.5 Income inequality

Rising income inequality can lower r* through reduced consumption and increased desired savings (Auclert & Rognlie, 2018; Rachel & Smith, 2017). Climate change is likely to increase income inequality and so reduce r* through these channels. New evidence suggests that rising income inequality might be a more important factor in explaining the decline in r* than demographics. Based on micro data, Mian et al. (2021a) weigh the relative importance of demographic shifts versus rising income inequality on saving behavior in the US from 1950 to 2019. These findings challenge the view that the aging of the baby boomers explains the decline in r* and might taper off as baby-boomers retire. Garbinti et al. (2018) report similar inequality trends in France for the after-war period. Albers et al. (2022) analyze the wealth distribution in Germany where, despite a tax policy regime aimed at reducing inequality, the top 10% of households increased their share of national wealth significantly over the past 25 years. To quantify the contributions of different channels that lower r*, Platzer & Peruffo (2022) use a heterogeneous agents overlapping generations (OLG) model with nonhomothetic preferences controlling for changes in intra- and intergenerational inequality. They find that the effect of rising inequality had dampened the natural rate as much as lower productivity and demographic changes.

The literature that identifies inequality as a driver of r* does not yet include climate change-related risks in the analyses. However, the impact could be substantial, disproportionately for poorer countries and communities more affected by acute and chronic effects, thus exacerbating their economic vulnerabilities and increasing global income inequality. Climate change will also exacerbate inequality: the impact of high temperatures is nonlinear; poor countries specialized in sectors affected by climate change have less capacity to adapt (IMF, 2017). Some studies show that such impact could have permanent adverse economic growth effects (e.g., Dell et al., 2012). Other research shows that countries with higher income inequality tend to have higher levels of per capita GHG emissions (Islam & Winkel, 2017). This hints at a feedback effect from inequality to climate change. Summing up, the general finding in the literature is that rising income inequality has a downward effect on r*.

4.1 Classical dichotomy

The natural rate of interest cannot be observed, which is why empirical research uses model-based approximations of r*, which by nature are beset with great uncertainty (Beyer & Wieland, 2017). Most approximations rely on observed interest rates, as the short-term rate in the HLW model or long-term bond yields in time series approaches for estimating r*. For that reason, our analysis of the possible impact of climate change could also be broadly interpreted as the long-run impact on observed interest rates. Besides the serious measurement and modeling issues, the mere existence of r* is debated in the literature, as it roots in the neoclassical theory that real economic variables and nominal variables like money can be analyzed separately (i.e., the classical dichotomy, see for instance, Fields, 2015). In the classical dichotomy, money is neutral, affecting only the price level and not real variables. This view implies that monetary policy is not an effective instrument in influencing the economy. The loanable funds theory, which we linked to the DICE model in Section 3, rests on this assumption. This is a controversial theory also because it assumes that prices are fully flexible and agents have rational expectations. In mainstream Keynesians and monetarist theories, the classical dichotomy is rejected, at least in the short-run, because prices are sticky. This implies that changes in the money supply and interest rates affect aggregate demand and thus the real economy in the short-run. It is a key element in CBMs, like New Keynesian dynamic stochastic general equilibrium (DSGE) models, in which monetary policy only affects the transition to a new steady state after a shock. In this view, monetary policy only has temporary real effects, but does very little to affect the trend itself (Woodford, 2003). It implies that monetary policy is not regarded as a driver of r*, since in the long run all prices and wages are assumed to adjust and the economy reverts back to its natural equilibrium (Gali, 2008).

4.2 Endogenous monetary policy

An implication of the New Keynesian theory is that monetary policy does not determine r*, given that this variable reflects the equilibrium between supply and demand in the long-term. From this, it follows that r* can be used as a yardstick for monetary policy, to determine the extent to which the policy interest rate should be reduced below r* in a recession and vice versa in a boom. However, this New Keynesian view is likewise debated. Recent studies have argued that monetary policy can have very long-term effects, such as on expectations (Hanson et al., 2015) and that r* might be endogenous to monetary policy (Borio et al., 2019; Gopinath et al., 2017; Jordà et al., 2020; Rungcharoenkitkul & Winkler, 2021). This refers to the potential impact of monetary policy on productivity growth, via hysteresis effects and the impact of the real interest rate on the allocation of capital. Fields (2015) summarizes this line of thinking by labeling r* as an conventionally determined exogenous distributive variable, which is strictly a monetary phenomenon.

Apart from the large uncertainty to measure r*, there are also fundamental reasons to assume that r* is not an obvious concept to guide monetary policy. Monetary policy aims at influencing the economy at the aggregate level, but according to Massonet (2015), the physical heterogeneity of capital goods prevents the determination of a unique rate of profit on the market for loanable funds. It implies that r* is unspecified in an economy with multiple goods, which makes it less useful as a guide for monetary policy decisions.

4.3 Monetary policy and climate change

There is a rapidly growing literature indicating that monetary policy has an impact on climate change and that this should also be taken into account in monetary policy considerations (for a recent and critical assessment of this literature, see Oman et al., 2022). In Section 3, we identified several channels for this, resting on the assumption that monetary policy may not be neutral in the long-run. In particular, if monetary policy affects economic trend growth (g), it has an impact on climate change via the feedback effects of potential output on carbon emissions (such feedback effects are part of CEMs such as DICE, as seen in Figure 2). Another channel through which monetary policy could have an impact on climate change is the influence on capital allocation. Prolonged accommodative monetary policy can slow-down the energy transition, if higher polluting energy-intensive industries benefit relatively more from monetary policy stimulus (Schoenmaker, 2021). This was, for instance, the case with asset purchase programs of the ECB, which were based on the concept of market neutrality. Market neutrality is derived from the Eurosystem's commitment to an open market economy which aims at minimizing the impact of monetary policy on relative prices and limiting distortions in market liquidity to preserve the price discovery mechanism of markets. It implies that the allocation of asset purchases follows capital market benchmarks, which are dominated by more polluting firms, because these are relatively more capital intensive. This carbon bias in the purchases of corporate bonds may imply that monetary policy does not achieve efficient outcomes (Schnabel, 2020). Ultimately this bias can reduce productivity in the economy and so (through parameter g) also contribute to lower r*, which will then be endogenous to monetary policy via climate change in the long run.

Oman et al. (2022) identify four ways in which monetary policy can be revisited to mitigate climate change, which they label (1) the standard economic approach, (2) the risk-based approach, (3) the proactive approach, and (4) the evolutionary approach. The standard economic approach views climate change primarily as a problem of negative externalities that have to be dealt with by other branches of government than the central bank. The risk-based approach recognizes the potential risk that climate change poses to the objectives of the central bank (price stability and financial stability) and its ability to reduce the effectiveness of central banks' policy instruments. These two approaches are currently the dominant views among monetary policy makers. A shortcoming of the risk-based approach could be that it assumes climate risks to be exogenous, but according to the concept of “double materiality” climate change not only affects financial institutions, but the financial system itself can also contribute to climate change and, hence, aggravate the very risks it wants to manage. The proactive approach wants to steer the economy toward low-carbon activities (and potentially other ecologically related goals), since problems with measuring climate-related financial risks, financial market inefficiencies (see Stern & Stiglitz, 2021), and the endogeneity of climate-related financial risks make purely market-based mechanisms less reliable. Finally, the evolutionary approach rests on the assumption that central banks' actions can only be understood in the context of broader institutional changes.

While a proactive approach is controversial and poses the risk of overburdening central banks' existing mandates (Bolton et al., 2020), some central banks recently more actively take into account climate considerations in their monetary policy. For instance, the ECB has adjusted, according to its Climate Action Plan, the framework guiding the allocation of corporate bond purchases to incorporate climate change criteria, in line with its mandate (ECB, 2021c). In October 2022, the ECB started directing its corporate bond purchases towards issuers with better climate performances as measured by diverse metrices: for example, the weighted average carbon intensity (WACI) and carbon footprint. These operational changes represent a form of risk management to protect the central bank's balance sheet from materializing transition and climate risks, different from outright “green” quantitative easing in the sense of Dafermos et al. (2018). A first report of the Eurosystem's climate-related financial disclosures on its corporate securities holdings for the years between 2018 and 2022 was published in March 2023 (ECB, 2023). It shows a decline in the respective metrices measuring the climate performance of assets, while at the same time an increase in the overall carbon emissions since the portfolio holdings increased during the pandemic. Importantly, the data coverage also improved over the 5 years, as an increasing number of bond issuers disclosed their carbon emissions.

The evolutionary approach of Oman et al. (2022) can also be applied in central banking, by using the relations in the financial sector to foster long-termism in financial activities (e.g., by stress-testing exposures for climate risk) and by supporting and complementing the transition policies of fiscal authorities and regulators by monetary tools (Bolton et al., 2020). Concerted climate policies in an orderly transition scenario (NGFS, 2021) have the potential to transform the macroeconomic and financial environment. This offers a ground to argue that central banks should engage in proactively integrating climate change considerations in their policies (see Benmir & Roman, 2021; Ferrari & Nispi Landi, 2021). At the same time, Oman et al. (2022) acknowledge the fact that the political mandates of central banks limit their scope for action. Most industrialized countries have central bank mandates that focus predominantly on price stability and financial stability as policy objectives. However, in some developing countries, central banks are obliged by law to support public investment policies by steering credit towards politically designated sectors. This provides more opportunities for proactive climate policies by the central banks.

Monetary policy can support climate transition policies by creating accommodative financing conditions for sustainable firms and activities. These lower the costs of capital and support a reallocation of production factors from carbon intensive to green sectors and fostering the financing of innovative green projects. Such an allocation process will be stimulated by adequate carbon pricing. This reduces possible adverse climate-related effects of monetary policy on productivity and can turn the interaction between monetary policy and potential GDP growth in the desired direction. There have also been ideas to use particular monetary instruments to foster the energy transition (see ECB, 2021b, for an overview). Central bank loans to banks could for instance be structured to incentivize banks to invest in green activities. It has been proposed to design such a structure by green discount interest rate on central bank loans to banks that are used to support the energy transition (van Tilburg, 2023). Another instrument mentioned in the public debate is for the central bank to purchase green financial assets in the context of the asset purchase programs (QE). Each of these ideas has pros and cons and implies the risk that central banks overstretch their mandates (ECB, 2021b).

Summing up, the various effects of monetary policy on r* should not be overemphasized as they may cancel out over the cycle. Nevertheless, there are various channels through which monetary policy is not neutral with regard to climate change. Foremost, monetary policy can influence the greening of the economy through the effects of monetary policy on the allocation of capital. These effects also depend on governments' climate policies that determine relative prices by imposing carbon taxes.