Policy Research Working Paper 9943 Macroeconomic Consequences of Natural Disasters A Modeling Proposal and Application to Floods and Earthquakes in Turkey Stéphane Hallegatte Charl Jooste Florent McIsaac Macroeconomics, Trade and Investment Global Practice & Climate Change Group February 2022 Policy Research Working Paper 9943 Abstract Turkey is vulnerable to natural disasters that can generate with distinct processes for the reconstruction of public substantial damages to public and private sector infrastruc- and private assets. The results show that destroyed infra- ture capital. Earthquakes and floods are the most frequent structure capital makes the remaining non-infrastructure hazards today, and flood risks are expected to increase with capital less productive, which means that disasters reduce climate change. To ensure stability and growth and min- the total stock of capital, but also its productivity. The imize the welfare impact of these disasters, these shocks welfare impact of a disaster—proxied by the discounted need to be managed and accounted for in macro-fiscal and consumption loss—is found to increase non-linearly with monetary policy. To support this process, the World Bank direct asset losses. Macroeconomic responses reduce the Macrostructural Model is adapted to assess the macroeco- welfare impact of minor disasters but magnify it when direct nomic effects of natural (geophysical or climate-related) asset losses exceed the economy’s absorption capacity. The disasters. The macroeconomic model is extended on several welfare impact also depends on the pre-existing economic fronts: (1) a distinction is made between infrastructure and situation, the ability of the economy to reallocate resources non-infrastructure capital, with complementary or substi- toward reconstruction, and the response of the monetary tutability between the two categories; (2) the production policy. Appropriate macro-fiscal and monetary policies offer function is adjusted to account for short-term complemen- cost-effective opportunities to mitigate the welfare impact tarity across capital assets; (3) the reconstruction process is of major disasters. modeled in a way that accounts for post-disaster constraints, This paper is a product of the Macroeconomics, Trade and Investment Global Practicea and the Climage Change Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http:// www.worldbank.org/prwp. The authors may be contacted at at shallegatte@worldbank.org, cjooste@worldbank.org, and fmcisaac@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Macroeconomic Consequences of Natural Disasters: A Modeling Proposal and Application to Floods and Earthquakes in Turkey* ephane Hallegatte1, Charl Jooste2, and Florent McIsaac3 St´ 1,2,3 World Bank Keywords: macro-structural model; natural disasters; earthquakes; floods; climate. JEL Classification: C10; C50; E52; Q54. * We thank Hans Anand Beck, Somik Lall, Miles Parker, and Christian Schoder for their helpful comments on an earlier version of this article. All remaining errors and opinions are our own. 1 Introduction All countries are exposed to various and varying natural hazards, either geophysical like earthquakes or climate-related such as floods, heatwaves, or windstorms. Increase in pop- ulation, economic growth, and urbanization are associated with a rapid increase in the economic damages caused by disasters. In the future, the increase in global average tem- perature induced by climate change is expected to increase the intensity and frequency of extreme weather and climate events further (IPCC, 2012), possibly leading to larger losses. Beyond physical damages — in the form of destroyed houses, roads, or factories — disas- ters also affect the functioning of the economic systems, with complex effects transmitted through supply chains and macroeconomic feedback. This paper proposes a model to in- vestigate and quantify the transition channels of disasters through the capital structure in a consistent macroeconomic framework. The model is applied to flood and earthquake risks in Turkey. This paper describes a methodological approach, extending Burns et al. (2019); Burns and Jooste (2019), to model physical capital stocks more granularly using Turkey as a case study. Macroeconomic models are used by finance ministries or central banks to quantify the effects of shocks and provide alternative paths for the economy. However, macroeco- nomic models have not been designed to capture specific impacts of natural disasters, such as the large share of damages affecting infrastructure and buildings (compared with other capital assets), or the practical constraints slowing down a reconstruction process (Halle- gatte and Vogt-Schilb, 2019). The main objective of this paper is to capture the structure and dynamics of natural shocks to capital and their link to economic losses and economic decision making. The modeling approach suggested in this paper could be applied to a wide range of disasters, provided they are channeled through the same mechanisms as those considered here for floods and earthquakes. It can also be used to explore how an increase in the frequency and intensity of hazards could translate into larger physical losses, and eventually larger macroeconomic impacts (including GDP, consumption, public deficit, public debt, inflation, etc.). And it could help explore how various macroeconomic or monetary policies can reduce macroeconomic (and welfare) losses, even if physical dam- ages remain unchanged. To incorporate disasters and mitigation strategies, the model architecture includes a segmentation of capital between infrastructure and non-infrastructure capital, and within this asset class, between the private and public sectors. The cost and return on capital are calculated using a nested structure with a constant elasticity of substitution. This paper can be viewed as an extension of Burns et al. (2020) along three lines. First, capital is divided into several dimensions and mapped to time series data. Second, the substitutabil- ity between infrastructure and non-infrastructure capital is estimated. This elasticity will therefore amplify or attenuate the damage, depending on the degree of substitution. Third, the modeling of post-disaster reconstruction takes into account the difference between pri- vate and public reconstruction and the different damages they face. Earthquakes and floods are the most frequent disasters in Turkey (see fig. 1). And according to data reported in Ocal ¨ (2019), earthquakes are the most destructive: they account for 92.6% of the number of damaged buildings and 95.9% of the number of demolished buildings in Turkey between 1900 and 2018. In this paper, we explore the 2 macroeconomic impacts of building losses due to earthquakes and floods. Our approach uses readily available capital data compiled by the World Bank1 and the Penn World Tables Feenstra et al. (2015). Earthquake risk data have been provided by the Global Earthquake Model, using the 2013 Euro-Mediterranean seismic risk model. Incidents 90 80 70 60 50 40 30 20 10 0 Earthquake Flood Epidemic Storm Landslide Wildfire Extreme Mass temperature movement (dry) Figure 1: Natural disaster incidents Source: WB Climate Change Knowledge Portal https://climateknowledgeportal. worldbank.org/country/turkey/vulnerability. 1.1 Related literature Conceptually, the analysis of natural disasters is informed by a vast literature in macroeco- nomics on the effects of uninsurable risks on economic growth and welfare (e.g., Aiyagari, 1994; Bewley, 1977; Krebs, 2003a,b). Natural disasters, as well as the mere possibility that they occur, can affect the macroeconomic situation in complex and important manners. For instance, it is long known that changes in economic risk can affect growth and welfare in interesting and unintuitive ways (see, e.g., Howell et al., 1993). In recent years, several theoretical studies have considered natural disaster risks and growth in particular (e.g., Akao and Sakamoto, 2018; Ikefuji and Horii, 2012; M¨ uller- urstenberger and Schumacher, 2015). F¨ One important finding in both empirical and theoretical literature is that the impact of disaster-related capital losses on output levels appears persistent, in contrast with what a simple Ramsey- or Solow-type model may suggest (e.g., Elliott et al., 2015; Hsiang and Jina, 2014; Raddatz, 2007; Strobl, 2011). Also, impacts of natural disasters on GDP tend to be nonlinear. Doubling the direct damages does more than doubling the GDP impact, in part because larger disasters delay the reconstruction process (e.g., Felbermayr and Gr¨oschl, 2014; Hallegatte et al., 2008). 1 https://www.worldbank.org/en/publication/macro-poverty-outlook 3 The literature also stresses the importance of considering differences in substitution options over the short-term (compared with long-term substitutability). Physical damages from disasters interact with the economy in a complex manner, through the network of supply chains (Baqaee and Farhi, 2020; Barrot and Sauvagnat, 2016; Boehm et al., 2019; Henriet et al., 2012; Inoue and Todo, 2019) and the role of infrastructure systems, from transport systems to electricity grids (Colon et al., 2021; Rose and Liao, 2005; Rose and Wei, 2013). In particular, destroyed infrastructure capital can make the remaining non- infrastructure capital unproductive, magnifying output losses (Hallegatte and Vogt-Schilb, 2019). To capture some of these effects, Hallegatte et al. (2007) introduce a specific produc- tion function to reproduce the short-term impact of disasters, accounting for the short timescales that do not allow (1) the same substitution among production factors as over the long-term, and (2) the reallocation of post-disaster assets to their most productive uses. The model also includes constraints on the reconstruction-related investments to make the model dynamics more consistent with observed disaster aftermath. With those changes, the model suggests that the GDP impact of disasters remain limited unless their frequency and intensity exceed the ability of the economy to reconstruct (in which case disasters start to show a cumulative effects that can significantly reduce the GDP level). The model also finds that disaster impacts are context-dependent, particularly in respect to the pre-existing macroeconomic situation, with disasters occurring during recessions having smaller impacts on GDP, as the reconstruction process can mobilize idle resources without crowding out other investments (Hallegatte and Ghil, 2008). Our work extends this work by representing the capital stock through different categories (infrastructure and non-infrastructure, public and private) to better represent the effect of capital losses on outputs, and real-world constraint on reconstruction investments. Our work also builds on recent papers on the interaction of natural risk and macroe- conomic factors. Isore and Szczerbowicz (2017) show that standard real business cycle models produce puzzling results when modeling disasters. They produce an increase in consumption when output and asset values fall in response to disaster risk. In this setup, the risk-free rate also falls in response to the disaster risk. To solve this puzzle, Isore and Szczerbowicz (2017) include sticky prices and an elasticity of substitution less than one. In their model, the discount rate varies over time and is a function of time-varying disas- ter risk (thus incorporating the role of preferences and uncertainty). This configuration generates an increase in patience and thus an increase in savings when the elasticity of substitution is less than one. If the elasticity of substitution is greater than one, this will result in an increase in impatience and an increase in consumption in response to a change in disaster risk. Another important feature of this model is the role of price flexibility. If consumers save in the presence of disaster risk, then investment will increase and result in higher output when prices are flexible. To counteract this result, sticky prices are required so that the demand for capital decreases with disaster risk and thus generates lower invest- ment. Finally, the risk premium will now increase in response to disaster risk, with an even larger response if agents are risk averse. Through careful calibration, this modification of the standard real business cycle model is able to produce a decline in output as well as a decline in consumption and investment with the change in the probability of an event. Another dynamic stochastic general equilibrium (DSGE) modeling approach by Wright and Borda (2016) represents disasters by adjusting the capital stock. It also includes the 4 country risk gap as a function of total factor productivity and the disaster shock. This modeling approach does not incorporate the fixed wage and discount rate characteristics of Isore and Szczerbowicz (2017). In this configuration, consumption increases for many of the countries while output and investment decrease, again producing the conundrum highlighted above. Camacho and Sun (2017) extend a Ramsey model to account for reconstruction times after a natural disaster, in this case earthquakes, and resilience capital to describe the economic impact in Japan and Italy. The numerical exercise shows that a large share of prevention capital (e.g., in a country like Japan with high resilience standards) reduces earthquake damage significantly for large shocks. However, the losses can be considerable for intense earthquakes. In the model, consumption absorbs most of the losses when re- silience capital is low. However, the share of consumption in GDP increases when adaptive capital is low in earthquakes. This is consistent with the DSGE literature regarding the optimal consumption smoothing scheme. Marto et al. (2018) build a dynamic small open economy model with permanent dam- ages to both public and private capital. Disaster leads to temporary losses in productivity, inefficiencies during the rebuilding process, and damage to the sovereign’s creditworthi- ness. In the model, investment in resilient infrastructure can be useful, especially if it is considered complementary to standard infrastructure, because it increases the marginal product of private capital, attracting private investment, while helping to withstand the impact of the natural disaster. The modeling framework in our paper shares many similar characteristics. Bakkensen and Barrage (2021) reconciles empirical and structural methodologies to model the economic impact of natural disasters. Instead of running reduced-form regres- sions of natural disasters on output per person, they identify the direct impacts on the structural determinants of growth (e.g., capital, labor, and total factor productivity). Their modeling approach has three important features: (i) separation of the effect of risk from the effect of a disaster; (ii) endogenous adaptation through changes in investment and sav- ings; and (iii) computing welfare costs associated with weather and climate risks. While they find that welfare is reduced by climate shocks (in this case cyclones), the impact on economic growth is smaller, in part due to the impact of risk (even unrealized) on precau- tionary savings. Finally, our analysis is part of a parallel literature on chronic physical risk, climate change, and economic growth. There is a growing body of literature that contributes to the active discussion on the evaluation of economic damage functions (e.g., Burke et al., 2015; Nordhaus and Moffat, 2017). With respect to incorporating climate disasters into macroeconomic models, some pioneering analysis of Fankhauser and Tol (2005); Moore and Diaz (2015) incorporate results from Dell et al. (2012) into the DICE model that had its origin in Nordhaus (1992). Because the estimates of the impact on output growth do not provide a clear match with the macroeconomic models, Moore and Diaz (2015) consider calibrating capital depreciation (and/or TFP growth) to match the reduced-form estimates. The choice between the two appears quantitatively significant in terms of optimal mitiga- tion dynamic. A similar question arises in Dietz and Stern (2015); Fankhauser and Tol (2005), which extend DICE to a long-run endogenous growth framework with capital- or investment-based knowledge spillovers. These papers show how the channeling of dam- ages, whether in consumption or investment, and the macro-modeling framework chosen, 5 are critical to assessing the impact of climate disasters. The modeling framework in this paper attempts to shed light on these channels by reinforcing the empirical and theoretical arguments in disaster modeling, whether natural or climate induced. 1.2 Natural disaster capital modeling considerations First, we adapt the physical capital structure of the World Bank’s macrostructural model MFMod (Burns et al., 2019) to capture the transmission channels of natural disasters in the macroeconomic framework. We assume that the aggregate capital stock is a combi- nation of two types of capital, infrastructure and non-infrastructure, and distinguish the contributions of private and public investments to the capital stock. Second, we use granular earthquake and flood damage data, and translate aggregate impacts into impacts to the various types of capital in the model. This allows to explore the importance of the substitutability between different types of capital, as opposed to assuming full substitutability (like in models where the capital stock is represented only by an aggregate value). Differentiated impacts across capital types allows us to represent the fact that disasters reduce the stock of capital, but also lead to the mis-allocation of the remaining capital, thereby reducing overall productivity (as discussed in Hallegatte et al. (2007); Hallegatte and Vogt-Schilb (2019) and empirically observed by Bakkensen and Barrage (2021) and Dieppe et al. (2020)). Finally, we take into account realistic reconstruction times, which in reality are con- straints faced by institutional, technological deficiencies, and a lack of financial access. The distinction between public and private capital also allows us to account for different decision-making processes, sources of financing, and constraints on timing for rebuilding these different asset types. The modeling of disasters require several data sets and methodological approaches. Figure 2 summarizes the key elements related to disasters, which include the identification of hazards, vulnerabilities and exposures. 6 Figure 2: Hazard, exposure, asset vulnerability, and socioeconomic resilience Source: WB https://www.worldbank.org/en/news/feature/2021/07/20/ in-europe-and-central-asia-the-poor-lose-more-when-disaster-strikes. We argue that these elements of disasters should be included in a macroeconomic model. First, possible hazards are best represented by their frequency and intensity. If all pos- sible hazards can be identified (which is challenging, especially for the most intense), then a probabilistic approach can be used to assess various statistics like the average annual loss, the likelihood of exceeding certain thresholds in losses (e.g., a drop in GDP larger than 1%), or the losses corresponding to various return periods (of annual probability of occurrence).2 Second, the exposure, i.e., the population and assets that are affected, should be iden- tified. In this case a link between the probability distribution of disasters to economic channels needs to be identified (in this paper the focus is on capital stock). This link can be enhanced with detailed data by region, e.g., the epicenter of the possible earthquake or flood maps, and the exposed assets and vulnerable people. Third, asset vulnerability needs to be estimated, possibly for various asset categories and different intensity levels. Vulnerability curves provide estimates of the likelihood that assets will be destroyed when exposed to disasters, or estimates of the repair costs (often expressed in fraction of complete reconstruction). These three elements provide an assessment of the physical damages caused by a nat- ural hazard. But the final economic cost — and welfare losses — depend on many other factors, and especially the ability of the affected economy to cope with and recover from the physical damages (what is referred to as ”socioeconomic resilience” in Hallegatte et al., 2017). Particularly important for the recovery process duration, and therefore total cumu- lative GDP or consumption losses, are the reconstruction investment choices that need to be modeled appropriately, taking into the account real-world constraints. The response of 2 The impact of disasters can also be explored using a scenario-based analysis (or ”what-if” scenarios), such as a category-5 hurricane or a magnitude-8 earthquake affecting a specific location. 7 government expenditures to natural disasters may depend on initial fiscal positions (i.e., the ability to finance additional expenditures). The private sector investment channel will depend on the returns and costs of capital after damages, but also on access to financing and self-financing capacity. Households’ consumption choices also matter. For instance, the constant coefficient of relative risk aversion implies that an increase in expected real rates will decrease consumption. Consumption may further deteriorate if the natural disaster causes permanent income losses. Finally, the response of the monetary and macroeconomic systems will affect the re- covery and total cost of a disaster. For example, a shock can affect prices through abrupt change in supply and thus increase consumer prices. Monetary policy may respond by rais- ing rates, if price stability is a country’s primary concern, thereby aggravating the economic shock. Monetary policy may also delay its response, as the shock is a supply shock (and not a typical demand shock).3 The section below explains how these features are incorporated into the World Bank’s macrostructural model. The paper is organized as follows: Section 2 describes the analyt- ical changes to the capital stock, government and private sector investment choices, and damage modeling. Section 3 describes the data used for damage and capital distinction. Section 4 presents several simulation exercises. Section 5 concludes. 2 Macro-modeling of natural disasters The key aggregate equations, long-run optimal behaviors, and accounting identities are based on Burns et al. (2019). We focus the modeling on the main contributions of this paper: the description of differential capital damage and the mapping of the data to theo- retical functional forms. We start from Cobb-Douglas aggregate production technology of final good,4 1−α Yt = At Ktα −1 Nt . Costs of production equal to the sum of labor costs and rental capital costs, Wt Yt = Nt + Rt Kt−1 . Pt where Yt is output, At is the total factor productivity, Kt−1 is end of period capital stock, and Nt is structural employment. α is the output elasticity, Rt is the real rental rate of capital, and Pt and Wt are prices of output and labor. Assuming cost-minimizing behavior, the (long-run) optimal aggregate capital to be rented equals its marginal productivity, ∂Yt∗ Yt∗ Yt∗ M P Kt = := Rt = α ⇒ Kt−1 = α . ∂Kt−1 Kt−1 Rt Note that at equilibrium, Yt∗ = Yt , or, potential GDP and actual GDP are equal to each other. Having defined the target (optimal) rental rate, we derive the components of capital out as CES functions and find the Rt that is consistent with the above. 3 See the empirical cases studied by Klomp (2020). 4 Variable names are borrowed from Burns et al. (2019), all other notations will be described in the text. The functional form need not be a Cobb-Douglas technology, but is used for ease of exploration. 8 2.1 Capital stocks Total capital stock is the sum of infrastructure capital (denoted with subscript S ) and non- infrastructure capital (denoted with subscript N ). Infrastructure and non-infrastructure capital is derived from both private (denoted with subscript P ) and public capital (denoted with subscript G). Given the equations above, we can derive solutions for both infrastruc- ture and non-infrastructure capital as well as a solution for the marginal product of capital that is consistent with aggregate capital. Figure 3 presents our main schema. Figure 3: Schematic representation of the modeling of capital We assume that capital is bundled together using a constant elasticity of substitution assumption (CES), with several nests. This allows us to estimate the complementarity or substitutability of private and public capital in the provision of infrastructure and other cap- ital. This addition makes an important contribution - capital damages in infrastructure as an example may also make non-infrastructure capital inoperable (see examples discussed in Burns et al., 2020; Hallegatte and Vogt-Schilb, 2019). As an example, the CES calibration for infrastructure and other capital could ensure (a behavior close to) complementarities, while the nest for private and government capital could result in substitutes. We assume that the economy is endowed with the following nested CES technology for the aggregate capital, 1 ρ1 ρ1 ρ1 Kt = ω1 KS,t + ω2 KN,t , for infrastructure, 1 P ρ2 G ρ2 ρ2 KS,t = ω3 KS,t + ω4 KS,t , 9 and non-infrastructure 1 P ρ3 G ρ3 ρ3 KN,t = ω5 KN,t + ω6 KN,t . where the share parameters for each capital (k ) is denoted with ωi and the elasticity of sub- 1 stitution σ = 1− ρ is either estimated or calibrated. Assuming that the final good producer minimizes its capital costs given the technology above, yields P P G G P P G G Rt Kt−1 = RS,t KS,t−1 + RS,t KS,t−1 + RN,t KN,t−1 + RN,t KN,t−1 , =RS,t KS,t−1 =RN,t KN,t−1 where the definitions of rental rates, Rs, follow the same logic as for aggregate capital and hence imply the equilibrium rates. As is standard in the literature, the first order conditions for infrastructure capital stock yield ∂Kt−1 1 ρ1 −1 : RS,t = λt Kt−1 1−ρ1 ρ1 ω1 KS,t−1 , ∂KS,t−1 ρ1 with λ is a Langrangian. The rental rate for aggregate infrastructure becomes ρ1 −1 1−ρ1 ρ1 −1 KS,t−1 RS,t = Kt−1 ω1 KS,t−1 = λt ω1 . Kt−1 Substituting the shadow cost value for marginal productivity of capital, i.e., λt = Rt , and solving for the optimal infrastructure capital stock KS,t yields, σ1 σ1 Rt KS,t−1 = ω1 Kt−1 , RS,t Similarly, we solve for non-infrastructure capital, σ1 σ1 Rt KN,t−1 = ω2 Kt−1 . RN,t In the same way we can write out the private and public optimal demands for both infrastructure and non-infrastructure capital, σ2 P σ2 RS,t KS,t−1 = ω3 P KS,t−1 , RS,t σ2 G σ2 RS,t KS,t−1 = ω4 G KS,t−1 , RS,t σ3 P σ3 RN,t KN,t−1 = ω5 P KN,t−1 , RN,t σ3 G σ3 RN,t KN,t−1 = ω6 G KN,t−1 . RN,t 10 The CES aggregator yields the aggregate optimal rental rate indices: 1 σ1 Rt = ω1 RN,t 1−σ1 + ω2 σ1 RS,t 1−σ1 1−σ1 , 1 σ2 P 1−σ2 σ2 G 1−σ2 1−σ2 RS,t = ω3 RS,t + ω4 RS,t , 1 σ3 P 1−σ3 σ3 G 1−σ3 1−σ3 RN,t = ω5 RN,t + ω6 RN,t . Following Burns et al.’s (2019) modeling choices, each capital rental rate should be equal to its own long-term replacement cost. However, in the short run, the premium and/or subsidies may differ due to certain rigidities. Therefore, the replacement cost is defined as i B PtI (rt + δt − πt + premt ) ∂Kt−1 ∂Kj,t−1 Uj,t = CIT × × i , Pt (1 − τt ) ∂Kj,t−1 ∂Kj,t −1 where i ∈ {P, G} and j ∈ {N, S }, δ is the capital depreciation rate, PtI is the price of B investment and Pt is the domestic price, rt is the risk-free rate (in this case the average 5 yield on government debt), premt is the risk-premium, πt is rate of inflation and τtCIT is the corporate tax rate. The following no-arbitrage condition must hold in the long run, i i Rj,t → Uj,t + ε, where the long-run level of the investment deflator converges to a constant ratio of the output deflator. In the short run, however, the growth rate of the investment deflator is a weighted average of nominal consumption inflation and its own lag, where the weight attached to past inflation, β , is estimated econometrically. Since there is no independent data on price deflators for public investment as distinct from private investment, nor for productive or adaptation investment, the four deflators are assumed to be equal ∆pi,j i,j C,XN t = α + θ pt−1 − pt−1 + β ∆pi,j C,XN t−1 + (1 − β )∆pt + εP I t . Note that all price indices are perfectly consistent with the aggregate capital stock from the production function and the different nests, as they have similar dynamics. 2.2 Investment decisions The investment choices of the private and public sector are considered next. Our starting assumption is that real capital stock for private infrastructure evolves ac- cording to a perpetual inventory method (hereafter; PIM), P P P P KS,t = (1 − δ )KS,t−1 + IS,t − IRt . Gross investments, I P , add to the existing capital stock net of depreciation, while recon- P struction investments, IRt are diverting a fraction of new investment. 5 Note that in Burns et al. (2019) the nominal interest rate was used as opposed to the average interest on debt. 11 Private investment decisions depend on (i) adjustment costs; (ii) expected returns and (iii) short-run returns vs. short-run costs. The framework is based on Tobin’s Q, where the Q ratio is equal to the return to capital relative to the cost of capital (or market value of assets to its replacement value). In this model, Tobin’s Q is defined as the ratio of the marginal product of capital to the cost of capital. The long run solves for the steady state ∗ investment-capital ratio, which equals potential GDP growth, yt , plus the rate of capital depreciation, δ . The standard empirical private investment equation is written as P P IS,t ∂Yt∗ P ∗ IS,t−1 P = β2 P − US,t −ε + (1 − β3 )(∆yt + δ ) + β3 P + εIP t N (1) KS,t −1 ∂KS,t−1 KS,t−2 where εIPt N is an iid residual. We will discuss the marginal productivity of private infrastructure capital shortly. Given that the modeling approach ensures that the investment variables, among others, are on the same balanced growth path, and given that capital stocks are technologically bound by the nested CES, it is easy to show that, in the long run, the following condition i i is satisfied ∀j Rj,t → Uj,t + ε. Therefore, in the long run, Yt∗ lim α = lim Ut t→+∞ Kt−1 t→+∞ T where Ut follows the same price aggregator as Rt . 2.3 Modeling of disaster channels Given that the theoretical design of natural disasters is the same as climate change when capital stocks are damaged, we use climate and natural disaster damages interchangeably henceforth. This section describes the aggregate damage functions for both public and private capital, and the channeling for disasters to the model. Residual and gross damages are expressed in terms of the cost of rebuilding the dam- aged capital. For example, if a disaster destroys a road and the cost of rebuilding the road is $1 million, then the official damage caused by the destruction of the road is $1 million minus the usual capital depreciation rate. Following Hallegatte and Vogt-Schilb’s (2019) main argument, simply subtracting the cost of rebuilding damage capital due to disasters from the total capital stock implies that the productivity of the destroyed capital was equal to the marginal product of the additional capital. In other words, capital remaining after the disaster can be reallocated instantly and without cost to its most productive use. In reality, climate damages include both infra- marginal and marginal capital, and make capital reallocation only partially possible (e.g, bridges cannot be moved) because it takes time and is costly. It is important to note that the productivity of infra-marginal capital is higher than that of marginal capital and therefore the expected output loss would be higher. This is in fact an extension of the observation that in the face of decreasing marginal productivity, average productivity will always be higher than marginal productivity. Assuming that the damaged capital is evenly distributed across sub-marginal projects, the economic value of the destroyed capital is equal to the average productivity of capital. 12 Capital destroyed by disasters has approximately average capital productivity, while newly constructed capital has marginal productivity. Following Hallegatte et al. (2007), to properly estimate the economic effects of climate damage it is necessary to keep track of un-repaired climate damage, DSt , and calculate its economic effect separately from new incremental capital projects. Building on Burns et al. (2020), the stock of un-repaired climate damage can be tracked using the following equation (we provide equations for the private sector, P , only, the public sector follows the same rationale), DStP = (1 − δ )DStP P P −1 + RDt − IRt , P where RDt is the residual damage (the gross damage net of the protection provided by the P adaptive investment) and IRt is the investment in repairing damages (past and present, indifferently) at time t, where repairs are equal to a constant share, ϕ, of the total invest- ment or total capital destroyed, whichever is smaller. During the reconstruction phase, substitution and reallocation of infrastructure assets are limited (Hallegatte and Vogt-Schilb, 2019) and we assume that the damaged capital alters potential real production as follows KLt−1 Yt∗ = −α At Ntα Kt1−1 , KS,t−1 :=Yt 1 P where KLt := ω3 (KS,t − DStP )ρ2 + ω4 (KS,t G − DStG )ρ2 ρ2 is the aggregation of both public and private though the same technology as capital. This functional form is chosen for the following properties: assuming that all capital is destroyed, there is no potential output. If half of the capital is destroyed proportionally between the private and the public, then po- tential output is halved. Note that when a natural disaster occurs, marginal productivities are affected, influencing investment choices. This framework is consistent with the intuitions of Hallegatte and Vogt-Schilb (2019) as reparation will always be preferred to building new capital (see Appendix 7.2 for the proof). Given that we have now defined real potential output, we are able to derive how dam- ages enter the investment decisions of the private sector. Specifically, we write the marginal product of private sector infrastructure capital   1−ρ2 1−ρ2 ∗ ∗ ∗ ∂Yt Y ∂Kt−1 ∂KS,t−1 Yt  KLt−1 Y KS,t−1 P =α t P + ω3 P P − t P . ∂KS,t−1 Kt−1 ∂KS,t−1 ∂KS,t−1 KS,t−1 KS,t − DSt Yt KS,t −1 Hence large damages to capital relative to output (weighted by the product of the aggre- gate infrastructure and private infrastructure capital share and the substitution elasticity) may generate large increases in the marginal product of the damaged capital, through DSt−1 and Yt∗ . Interestingly, non-infrastructure capital marginal productivity will take the form ∂Yt∗ Yt∗ ∂Kt−1 ∂KN,t−1 P =α P . ∂KN,t −1 Kt−1 ∂KN,t−1 ∂KS,t−1 Turning to investment in reconstruction, given that the capital destroyed by climate disasters has approximately an average capital productivity, it would be optimal to direct 13 all investment to reconstruction. However, agency problems (not all investors will own the damaged capital) and regional and sectoral capacity constraints will prevent this out- come from being observed. To reflect these considerations, it is assumed that rebuilding investment cannot exceed a ϕ(t) share of total investment, P IRt = min (1 − δ ) DStP P P −1 + RDt , ϕIS,t . (2) We assume that the ϕ is constant and equals 5%. Note that the model does not rule out total capital destruction. However, it is not realistic to assume that all capital will be destroyed even in the case of a massive disaster. We therefore assume that about 10% of the capital is indestructible, ˜ t) DSt = min(DSt−1 + RDt − ItR , 0.9K Further note that the model ignores the responses of labor productivity and human capital to natural disasters. The model may also underestimate the actual impact of natural disasters given the rebuilding time. 2.4 Data and estimation The macroeconomic model is estimated using the methodology presented in Burns and Jooste (2019). The production function technology is assumed to have a Cobb-Douglas form. Capital is aggregated using a CES function, with our main elasticity of interest being σ T , which mea- sures the substitution between infrastructure and non-infrastructure capital. This parame- ter is important for computing the changes to the marginal product of capital, and hence to the profitability of firms, to natural disasters. The productivity of non-infrastructure capital will rise or fall in line with an increase or decrease in infrastructure productivity if the two capital variables are complements. To estimate this parameter, we collect data on investment and capital stock from several sources. The World Bank produces aggregate data on investment, private investment, and public investment for several countries in the Macro Poverty Outlook. The global PENN tables, discussed above, also include data on capital based on structures and non-structures and their prices, which allows for a rough estimate of the elasticity of substitution. Using panel estimation with standard fixed effects, we estimate a simple elasticity of substitution using data on the ratio of infrastructure capital to non-infrastructure capi- tal as a function of the relative price of non-infrastructure capital to infrastructure capi- KS tal, ln K N ≈ αi + σ T ln U US N . The elasticity is statistically significant and equals σ T = 0.58(0.03), which implies that structures and non-infrastructure capital are complements. 1 Indeed, given that σ T = 1− ρ1 , if ρ1 → −∞ then the function approaches a Leontief produc- tion function. Unfortunately, data availability does not allow us to estimate the elasticity of substitu- tion between private and public infrastructure or between private and public capital outside infrastructure. We assume that they are substitutes and set σ S (ρ2 ) = σ N (ρ3 ) = 2.6 6 For other estimates of aggregate capital substitution between the private and government sector, see An et al. (2019). 14 3 Macroeconomic impact of a natural disaster We start by generating a model-determined baseline without shocks for 200 periods, i.e., 200 years, that we identify as the steady state.7 We then proceed to introduce shocks relative to the steady state baseline. This setting allows us to reduce the influence of short- term deviation impacts from the balance growth path for the shock analysis. In this section, we compare results for different disaster sizes, ranging from 1% to 20% loss of the infrastructure. While the largest disaster is perhaps unrealistically large, it illustrates several useful properties of the model. We also reproduce a disaster of a magnitude similar to the 1999 Marmara Earthquake, as one example of large plausible shock. The next set of simulations draw from the empirical distribution using the GEM data to extract uncertainty intervals for the economic outcomes and to assess potential losses from realized data. 3.1 Economic response after a natural disaster Throughout the simulations, results are displayed relative to the baseline in which shocks and resilience strategies are absent from the analysis. The natural disaster shock is intro- duced in year three, discussed shortly. In the scenario showed in figure 4, the physical damages from the disasters are fully re- paired within three year. However, the consequences of the disasters are almost permanent in levels. The top left hand side panel shows that real GDP, consumption, and investment fall after the shock. The shock also generates inflation (top panel on the right), which degenerates back to the baseline level for a decade after the shock. The largest increase in inflation occurs after the shock. The inflation response is mainly due to a cost-push mechanism, following a co-mutual increase in the rental rate of capital and in the output gap (potential output falls faster than actual output). Real demand does fall, but less than potential, leading to an inflationary response. One may also view this in the context of factor prices. The rental rate of capital increases when capital is destroyed (supply of capi- tal falls relative to demand). The impact on wages is complicated given the existence of a wage-price loop. This modeling result can be compared with empirical findings on most natural disas- ters that report increased inflation after climate shocks (see e.g., Parker, 2018). However, empirical results for the case of earthquakes remain inconclusive. Cavallo et al. (2014) studies the impacts of the 2010 Chilean and the 2011 Japanese earthquakes and show that shortages at the aftermath of these events did not translate into higher prices. Moreover, Doyle and Noy (2015) showed that the Canterbury earthquake in 2010 generated a fall in aggregate demand that decreased prices (note that the results were insignificant). How- ever, the Canterbury earthquakes are not a great example for drawing wider conclusions since the rate of insurance was high (Wood et al., 2016). Furthermore, using global data, Parker (2018) shows that earthquakes increase the price of energy, clothing and footwear, while food prices seem to have declined in the first few quarters after the shock. More recently, using a simultaneous equation model with global data, Klomp (2020) finds a positive response of inflation after the occurrence of an earthquake. 7 See Figure 17 in the appendix. 15 Demand response Price response .1 .5 % pt. deviation from baseline % deviation from baseline .4 GDP .0 .3 HH. Cons -.1 .2 Fcst -.2 GDP .1 Imports -.3 HH. Cons Wages .0 -.4 Tot. Inv. -.1 -.5 Exports -.2 -.6 Imports -.3 0 4 8 0 4 8 12 16 20 24 28 32 36 40 44 48 52 12 16 20 24 28 32 36 40 44 48 52 Cost of capital and MPK responses Fiscal and trade .4 .100 % level dev. from baseline % deviation from baseline .3 .075 Aggregate MPK .050 .2 Aggregate UCC .025 .1 .000 .0 -.025 -.050 Debt(%GDP) -.1 -.075 Trade(%GDP) -.2 -.100 0 4 8 0 4 8 12 16 20 24 28 32 36 40 44 48 52 12 16 20 24 28 32 36 40 44 48 52 Monetary policy Growth .4 .6 % level dev. from baseline % deviation from baseline .3 .4 Monetary policy rate .2 .2 Lending rate .0 .1 -.2 .0 -.4 GDP -.1 -.6 Potential -.2 -.8 0 4 8 0 4 8 12 16 20 24 28 32 36 40 44 48 52 12 16 20 24 28 32 36 40 44 48 52 Figure 4: Economic impact of a 1% degradation of infrastructure If no sovereign resources are devoted to recovery outside of reconstruction spending, then the fiscal response assumed in the paper will cause the aggregate gross debt-to-GDP ratio to return to its baseline value. However, it should be noted that the initial rise in prices leads to an increase in the nominal tax base (e.g., nominal GDP) as inflation exceeds the decline in real GDP. As a result, we see a reduction in the gross debt-to-GDP ratio (the level of nominal debt is higher compared to the baseline). Two important caveats should be mentioned. First, we assume that the expenditure functions are neutral, i.e. that other expenditures aim to achieve a long-term neutral budget deficit. Second, we assume that debt is below a level associated with default risk. In Turkey, the fiscal debt-to-GDP ratio is well below sustainability threshold reported in the literature at the time of writing. If this ratio were to rise above the sustainability threshold, approximately 60%, then the market premium would increase, diverting more resources to debt service. The results would then be reversed if spending increases at the aftermath of an earthquake. Moreover, because we model monetary policy as a Taylor rule, the nominal interest rate rises in response to inflation, triggered by the increase in the real marginal product of capital. The response of real private investment in infrastructure deserves particular attention. 16 Although the marginal product of capital is higher than the cost of capital in the medium- term, investment after a disaster is below the baseline. This is primarily driven by the investment equation (Eq. 1). The investment choices include the remaining capital, taking into account the reconstruction and adjustment investments of the previous period. The combination of lower expected real income and the inertia of the investment decision im- plies that the increase in investment due to the marginal product of capital is completely offset in subsequent periods. In turn, the government’s investment response (those not de- voted to reconstruction) is assumed to hit a fixed share of fiscal revenues to maintain bud- get neutrality (a modeling assumption not necessarily reflecting historical fiscal responses by Turkey). Taxes, and thus government spending decisions, are sensitive to movements in the tax base, which is itself sensitive to the degree of price rigidity. Sticky prices and active monetary policy can further reduce output after a natural disaster shock. Such a scenario may lead to a large increase in the user cost of capital (larger than the return to capital), which reduces the expected profits from investing in the short- to medium-term. We will further discuss the role that the monetary policy plays shortly. 3.2 Response to larger shocks To better explore the response of the model, we perform a sensitivity analysis, reproducing milder to more severe natural disasters. We also include a disaster of equivalent intensity to the 7.8 Richter scale earthquake that devastated northwest Turkey at 3:02 a.m. on August 17, 1999. Results are shown in figure 5. GDP responses Potential GDP responses 0 0 % deviation from baseline % deviation from baseline -1 -2 -2 -3 -4 -4 1% Damage -6 -5 4% Marmara 10% Damage -8 -6 -7 -10 0 3 6 9 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Inflation Current account balance % pt deviation from baseline (%GDP .4 % pt deviation from baseline 6 .2 4 .0 -.2 2 -.4 0 -.6 -2 -.8 0 3 6 9 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Figure 5: Sensitivity of economic response (GDP and inflation) to capital losses of different intensity, including a loss equivalent to the 1999 Marmara earthquake. The model used in this paper estimates that the 1999 Marmara earthquake resulted in a loss of 1.5 percent of GDP in the year of the event, which is much lower than the 4 percent decline in potential GDP. GDP losses increase to 2 percent of baseline GDP two years after the shock and then decline, with the rebuilding process completed in six years. 17 The earthquake occurred during an economic recession that makes it difficult to quantify empirically the GDP losses. For comparison, a model-based simulation done after the event (World Bank, 1999) suggested immediate losses amounting to 0.6 to 1 percent of GDP. These results show that the economic response is not linear: the model response is different for small and large shocks. This nonlinearity is, in part, determined by equation 2 as well as the interaction of the output gap on prices and interest rates. These responses of inflation illustrate this point, with a more complex dynamics for larger shocks. Two additional points are worth mentioning. First, while supply (i.e. potential GDP) reacts strongly to large shocks, economic mechanisms and dynamics smooth and delay the economic response, giving GDP responses their curvature. Second, the V-shaped response of potential GDP reflects the recovery process of investment in the economy. While it takes about three years to recover from a 1% loss in the capital stock, the recovery periods increase to six and fourteen years respectively for a Marmara-sized shock and a 10% shock. To better explore the investment dynamics, figures 6 and 7 show the results after a 20% damage to infrastructure capital. In this scenario, the dynamics is different, in part because reconstruction extends over a 20-year period. In this case, GDP falls by about 14% and it takes more than 50 years to return to the trend. The marginal product of capital increases by just over 5%, which induces some positive response from infrastructure investments (in addition to reconstruction investments). This large-scale destruction generates a surge of inflation, which will have an impact on household consumption through real income. While the debt-to-GDP ratio initially declines, mainly because the denominator is dominated by prices, it then increases when inflation returns to the baseline. Figures above show that aggregate private investment falls after a major disaster. At first glance, this may seem odd given the strong increase of the marginal product of capital, which should rise sharply after the shock. However, several factors come into play that lead to a reduction in private investment, especially in infrastructure. The bottom panel of figure 7 shows the stimulus spending after a large disaster (20% of capital). By far the most important response is from the private sector, which accounts for the largest share of total capital stock in Turkey. The top panel shows the deviation of investment from the baseline. The share of infrastructure investment in GDP, before ac- counting for reconstruction, initially increases after the shock and then deteriorates, while non-infrastructure investment increases. Figure 7 shows that the model takes about 20 years to recover from a natural disaster that destroys 20% of the infrastructure. Although it is difficult to compare this result with historical data, this length of the reconstruction period can be considered pessimistic for large shocks, probably because the policy response to such a large shock would be different from the behavior represented in the model.8 The dynamics of private sector investment are determined by three factors: the persis- tence of investment decisions, the level of capital stock, and the net return on capital. Figure 8 shows that the net return on capital is volatile in the short-term, responding to real and monetary factors. However, this volatility has little influence on capital stock. The medium- to long-term infrastructure decreasing response is primarily determined by the reduction in the amount of infrastructure capital relative to the baseline. To under- stand these medium- and long-run supply-side responses, we focus the interpretation on 8 This point will be further discussed in the next section. 18 Demand response Price response 4 20 % pt. deviation from baseline % deviation from baseline GDP 0 15 HH. Cons 10 Fcst -4 GDP 5 Imports -8 HH. Cons Wages Tot. Inv. 0 -12 Exports -5 Imports -10 -16 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Cost of capital and MPK responses Fiscal and trade 20 10 % level dev. from baseline % deviation from baseline 15 5 10 Aggregate MPK 0 Aggregate UCC 5 -5 0 -10 Debt(%GDP) Trade(%GDP) -5 -15 0 3 6 9 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Monetary policy Growth 20 4 % level dev. from baseline % deviation from baseline 16 0 -4 12 Monetary policy rate -8 8 Lending rate -12 4 -16 GDP 0 -20 Potential -4 -24 0 3 6 9 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Figure 6: Economic impact of a 20% degradation of infrastructure the denominator of the investment equations. The top panel of Figure 8 shows that both infrastructure and non-infrastructure capital decline following the shock, more pronounced for infrastructure. Since reconstruction investments are financed by equity in these simu- lations, resources are diverted away from productive capital investments that would have been made in the absence of shocks. The net effect of the shock on capital is negative. The dynamics of the positive response of non-infrastructure investment relative to GDP is explained by the fact that there is a strong inertia of this type of investment while GDP, the denominator, deteriorates. Thus, non-infrastructure investment’s share increases while its stock is lower in the scenario where the shock occurs. We note that the signs of the variations between infrastructure and non-infrastructure are similar throughout the simu- lation, which is the consequence of the complementary between these two types of capital. 19 Investment not related to reconstruction % GDP deviation from baseline 2.5 2.0 Infrastructure investment 1.5 Non-infrastructure investment 1.0 0.5 0.0 -0.5 -1.0 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Recovery investment % GDP deviation from baseline .8 .6 .4 .2 Private Government .0 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Figure 7: Investment responses to a large shock 4 The role of macroeconomic and monetary factors This section explores the importance of macroeconomic factors in determining the total impact on welfare after natural disaster shocks. It shows that modeling disasters in a macroeconomic framework allows to account for mechanisms that have a significant impact on welfare losses, but also allows for the evaluation of macrofiscal or monetary policies and options that can mitigate the impact of disasters. The section begins by discussing the relationship between asset losses and consumption (and welfare) losses. It then looks at the role of economic conditions (boom or bust) at the time of the disaster, and finally explores the role of monetary policy. The results presented below illustrate the importance of examining natural disasters in a macroeconomic context. First, because macroeconomic and monetary factors play an important role, making asset losses a poor indicator of the impact of a shock on welfare. Second, the macroeconomic framework opens up the possibility of cost-effective interven- tions to mitigate the welfare effects of disasters, including specific fiscal or monetary rules assumed in the paper. 4.1 From asset losses to consumption losses Figure 9 summarizes the consumption losses in net present value (NPV) for different mag- nitudes of the natural disaster damages, assuming a 6% discount rate. The x-axis represents the hypothetical damages on capital (in terms of GDP), while the y-axis summarizes the net present value response. For comparability, both are expressed in percentage of GDP after a disaster shock. 20 Infrastructure capital stock Non-infrastructure capital stock .01 % level dev. from baseline % level dev. from baseline 0 .00 -.01 -2 -.02 -4 -.03 -.04 -6 -.05 -8 -.06 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Infrastructure net return to capital Non-Infrastructure net return to capital .004 % level dev. from baseline % level dev. from baseline .16 .000 .08 -.004 -.008 .00 -.012 -.08 -.016 -.16 -.020 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Figure 8: Capital responses 50% 5% reconstruction 45% 20% reconstruction % NPV Consumption losses 40% 35% 30% 25% 20% 15% 10% 5% 0% 0% 5% 10% 15% 20% 25% Direct losses (% GDP) Figure 9: Direct versus the net present value consumption losses: Reconstruction lags Figure 9 shows that total consumption losses (discounted) increase nonlinearly with the amount of damage. For large disasters, total consumption losses exceed the value of the physical damages, especially when the monetary policy is not disaster-sensitive. In our simulations, consumption losses exceed asset losses when the latter are bigger than around 6% of GDP. For smaller shocks, these simulations suggest that the economy can absorb the shock, building on its ability to borrow and reallocate resources, so that the final consump- tion losses are simply equal to the replacement cost of the lost capital. For larger shocks, the longer duration of the reconstruction period and the propagation of impacts (with a reduction in the production of the non-affected capital) magnify consumption losses. Qualitatively, these results are consistent with empirical results (e.g., Felbermayr and Gr¨oschl, 2014) and previous modeling exercises (e.g., Hallegatte and Ghil, 2008), but real- ized within a more sophisticated macroeconomic framework. They confirm two important insights: (1) for large-scale disasters, consumption losses can exceed direct damages; (2) 21 macrofiscal and monetary policies can play in important role in mitigating the impact of large disasters. In the case of Turkey, our model suggests that the threshold level at which consumption losses exceed asset losses is high, and the nonlinearity is low. At 6% of GDP in the baseline scenario, this threshold is higher than the most severe floods and earthquakes that are most likely in the probability distributions considered in the next sections of the paper. This result should be viewed with caution, however, as this threshold will depend on a few parameters that are difficult to estimate, such as the economy’s ability to mobilize resources for reconstruction. Figure 9 illustrates this point. By increasing the capacity to mobilize (or divert) resources for reconstruction, changing the parameter ϕ in eq. 2 from 5% to 20%, the threshold increases from 6% to 21%. The faster the economy is able to rebuild, the less welfare is lost. 4.2 The importance of the economic situation before the shock The previous simulations were performed on a steady state growth path, but disasters affect economies at specific phases of their business cycle. For example, the Marmara earthquake affected Turkey during a recession. However, the economic response to a shock is not likely to be similar: in particular, the increase in demand created by reconstruction is less likely to displace workers and production capacity during a recession than during an expansion. In addition, the demand created by reconstruction can act as a stimulus, boosting aggregate demand and accelerating the macroeconomic recovery and reconstruction process. For example, the 1992 Hurricane Andrew hit Florida when half of the construction workers in the state were unemployed. We simulate the same disasters affecting the same economy, but during an expansion or recession phase of its business cycle. Figure 10 shows that the impact of the disaster is similar, but the drop in GDP is smaller and the recovery faster if the economy is in recession before the shock. These results confirm previous theoretical and empirical findings from Ginn (2021); Hallegatte and Ghil (2008) that economies in recession are more resilient to natural disasters. 4.3 The importance of monetary policy The role of monetary policy and the degree of price stickiness are also important factors that determine the total economic cost during disaster shocks. In the model, inflation is anchored by inflation expectations, while the dynamics of inflation depend on the degree of price stickiness when marginal costs change. Results in the previous section illustrate the classical monetary policy dilemma: how to accommodate the supply-side real shock without destabilizing inflation expectations? We conduct a monetary policy experiment to shed light on possible policy actions. In practice, we compare the model’s responses in cases with rigid and flexible prices, under an active monetary policy or a delayed monetary policy response that does not react immediately to the disaster-related response in inflation. Figure 11 shows that the price dynamics is dominated by changes in the marginal prod- uct of capital, which increases aggregate marginal cost. Active monetary policy does not 22 GDP Potential GDP 5.00 5.00 % deviation from boom % deviation from boom 0.00 0.00 -5.00 -5.00 -10.00 Potential GDP (Boom) -10.00 GDP (Boom) GDP (Recession) -15.00 Potential GDP (Recession) -15.00 -20.00 -20.00 -25.00 1 6 11 16 21 26 31 36 41 46 51 56 61 1 6 11 16 21 26 31 36 41 46 51 56 61 Output gap Inflation 16.00 20.00 % deviation from boom % deviation from boom 14.00 12.00 15.00 10.00 8.00 10.00 6.00 Output gap (Recession) Inflation (Recession) Output gap (Boom) 5.00 Inflation (Boom) 4.00 2.00 0.00 0.00 -2.00 -5.00 1 6 11 16 21 26 31 36 41 46 51 56 61 1 6 11 16 21 26 31 36 41 46 51 56 61 Figure 10: Responses during recessions vs. booms significantly reduce inflation after the shock, because the initial supply-side shock out- weighs the impact of monetary policy. Furthermore, active monetary policy reduces further consumption, investment, and thus output (see right panel), which in effect amplifies the recessionary effects after natural disasters. In a flexible price system, inflation returns to equilibrium faster, but results in a larger cumulative loss of output. Inflation 0.00 Output 25.00 -2.00 20.00 Delayed monetary policy (sticky prices) Delayed monetary policy (flexible -4.00 % pt. deviation from baseline % deviation from baseline 15.00 prices) Monetary policy (flexible prices) -6.00 Monetary policy (sticky prices) 10.00 -8.00 Output (sticky prices + delayed 5.00 monetary policy) -10.00 Output (flexible prices + delayed monetary policy) Output (flexible prices + monetary 0.00 policy) -12.00 Output (sticky prices + monetary policy) -5.00 -14.00 1 11 21 31 41 51 1 11 21 31 41 51 Figure 11: Responses conditional on monetary policy Delayed monetary policy response assumes a lag in the reaction function during the pe- riod of cost-push pressure immediately after the shock, supposing that the inflation created by the sudden supply-side shock should not trigger monetary tightening. This simulation would be equivalent to a catastrophe escape clause in the application of the Taylor rule. Figure 11 shows that the suspension in monetary policy response can mitigate the output 23 and consumption impact significantly, without generating large trade-off with long-term inflation. Our findings are aligned with past monetary policy practices reported in Klomp (2020) in the context of earthquakes. This paper shows that the policy interest rate drops in the first year after an earthquake, meaning that monetary authorities prioritized recovery over price stability. The plausible explanation provided by the author is that demand shocks are more likely to dominate supply shocks, which echoes the result of our simulations when we compare actual and potential GDP growth rates (see Figure 6). These results call for more research on the design of monetary policy in post-disaster and crisis situations, a topic with considerable uncertainty. As an example, Ehrmann and Smets (2003) explore the economic consequences when monetary policy reacts to incorrect information, and show that a more conservative approach (i.e., reducing the weight on the output gap in the interest rate reaction function) reduces the welfare loss from supply-side shocks. And Ferrero et al. (2019) show that the uncertainty on the inflation response to the output gap should lead to a monetary policy response that is more aggressive for transient shocks and more cautious for permanent shocks. Natural disasters are exogenous, short, and transitory shocks. They are therefore less likely to destabilize inflation expectations. In this particular case, our model suggests that a temporary suspension in monetary policy (i.e., not increasing rates immediately following a natural disaster) helps the economy cope with the transient supply-side shock. However, in the context of climate change, when particular shocks become increasingly frequent, they may affect inflation expectations and a similar strategy may have unintended adverse effects (Batten et al., 2016; Rudebusch et al., 2019). In the current Turkish context, it should be recognized that inflation is historically high and that inflation expectations may not be anchored and therefore sensitive to shocks under a loose monetary policy regime. To illustrate the potential benefit from a more effective monetary policy response, figure 12 shows the consumption losses in net present value (NPV) for different magnitudes of the natural disaster damages, but this time with the two monetary policy responses. If the monetary policy response is suspended for a short period after the shock (so that it responds only to second-round inflationary effects), the welfare impact of the disasters is reduced. In particular, threshold beyond which consumption losses exceed asset losses increases from 6% to roughly 11% (see orange curve), making the economic system significantly more resilient to the shock. 5 The cumulative economic impacts of earthquakes and floods in Turkey In this section, we shift from the impact of one disaster to the impact of a distribution of possible disasters, described by the probability of occurrence and intensity (expressed in asset losses). 5.1 Earthquake risk The seismic risk assessment and retrofit scenarios for Turkey are based on a Global earth- quake model (hereafter; GEM) available in Rao et al. (2022). Using exposure model de- 24 40% Without MP suspension 35% With MP suspension % NPV Consumption losses 30% 25% 20% 15% 10% 5% 0% 0% 5% 10% 15% 20% Direct losses (% GDP) Figure 12: Direct versus the net present value consumption losses, with and without monetary policy suspension velopment, probabilistic seismic risk analysis, and retrofit intervention scenarios, GEM pro- vided loss curves for sixteen building occupancies (e.g., government, industry, commerce, education, and healthcare). Assuming a mapping of public or private occupancies, we are able to derive the percentage loss of total infrastructure from the disaster effect for specific frequencies (1, 2, 5, 10, 20, 50, 100, 200, 250, and 500 years). Figure 13: Percentage of infrastructure loss by seismic frequency event in 2020 given the current stock of infrastructure Source: GEM. Note: The light colored lines represent the percentage losses by occu- pancy. The thick black line represents the aggregate for the private sector and the thick blue line represents the aggregate for the public sector. Figure 13 shows that the maximum percentage loss for a seismic event below the 100- year frequency is relatively small, less than 1%. Looking at the effect of the lowest fre- quency, the 500-year frequency, the losses for the entire occupancy are between 1.5% and 25 3.5%, with an average of about 1.8% for the public sector and 2.5% for the private sector. Since the private sector is much larger than the public sector in terms of buildings, the global average is close to that of the private sector. Two caveats are worth mentioning. First, the disaster coverage is for buildings only. Therefore, simulations based on these data will not provide impacts on infrastructure, for experiments with GEM infrastructure has to be understood as buildings. Second, these results are based on the 2013 Euro-Mediterranean seismic risk model. Early simulations of the 2020 Euro-Mediterranean seismic risk model for Turkey may suggest higher overall impacts. A Monte Carlo simulation draws from the empirical cumulative distribution function, −1 FX , using the empirical inverse probability integral transform, FX (u) = x.9 Our simula- tions start in 2021 as opposed to the steady state simulations earlier. Figures 14a and 14b summarize the stochastic responses for GDP and consumption, respectively, between 2021 and 2091. The median macroeconomic impact of earthquakes is small, with consumption losses much smaller than GDP losses. The distribution function has negative skewness and the tails are flat. This shape is due to the long-term impact of shocks in the system. Indeed, if the shocks had mostly a short- term impact, the distributions would have roughly the same shape over time. This result is consistent with empirical results suggesting that earthquakes have long-term impacts on GDP. For instance, Lackner (2018) uses earthquake data from 1973 to 2015 to estimate the long-term effects on GDP per capita and finds that an earthquake can reduce GDP per capita by up to 1.6% eight years later. (a) GDP (b) Consumption Figure 14: GDP and consumption responses to earthquakes 5.2 Flood risks, with and without climate change Flood damages are the second most frequent natural disaster in Turkey following earth- quakes. However, figure 15 shows that the direct capital losses from floods are quite small, even for rare events. For example, a 1 in 1500 year event destroys less than 1% of total capital. 9 Probabilities were provided in discrete intervals. To simulate the full distribution of the data, the points between the probabilities were interpolated using cubic splines. 26 Floods (% damage of capital) 0.70 0.60 0.50 % of capital 0.40 0.30 0.20 0.10 0.00 20 50 100 250 500 1000 1500 Annual returns Figure 15: Exposed asset value loss for different return periods Source: UNISDR (2015) and Myhre et al. (2019). The change in flood frequency due to climate change is taken from Myhre et al. (2019). The frequency of floods, F , is assumed to double with every degree in temperature rise T , consistently with basic physics: F (T ) = 2T −T0 × F0 . The temperature rise scenarios are based on RCP 2.6 and RCP 8.5. The first column of Table 1 reports the expected damage to the capital stock arising from floods of increasing severity. Column 2 shows the probability of each of those events under a 1°C of warming scenario. The columns for 2°C of warming and 3.7°C of warming show the increased probabilities based on the above formula. Damage Prob for +1°C Prob for 2°C Prob for +3.7°C 0 91.23% 75.20% 19.44% 0.08% 5.00% 14.14% 45.95% 0.14% 2.00% 5.66% 18.38% 0.19% 1.00% 2.83% 9.19% 0.34% 0.40% 1.13% 3.68% 0.42% 0.20% 0.57% 1.84% 0.52% 0.10% 0.28% 0.92% 0.65% 0.07% 0.19% 0.61% sum 100% 100% 100% Table 1: Current and estimated probabilities for flood damages of a given size Figure 16 shows the impact of floods on a set of macroeconomic variables, namely GDP, consumption, private investment, consumer prices, real consumer wage, and real producer wage. The simulations start in 2020 and are presented until 2100. Under RCP8.5, median economic losses in terms of GDP and relative to the baseline grow to 0.5% by 2100, in contrast to losses of slightly larger than 0.1% under the RCP2.6 scenario. 27 Floods reduce both consumption and investment, a similar result for the earthquake analysis. However, due to the small economic impact of the shock, factor prices (in this case wages) do not change materially from the baseline. In other terms, the scale of the floods expected in Turkey remains small enough to make the macroeconomic and monetary response negligible, and their frequency is small enough to ensure that they have limited cumulative impacts. This does not mean that floods do not represent a significant economic and welfare cost (especially on local economies), but only that their assessment does not require to be done in a macroeconomic framework. This result echoes results from World Bank (2021), who explored the impact of floods and droughts in Argentina, and concluded that floods represented a major threat to poor people, in spite of a limited macroeconomic aggregate impact. GDP, in percent compared to baseline Consumption , in percent compared to baseline .1 .1 .0 .0 -.1 -.1 -.2 Scenario: RCP: 2.6 Scenario: RCP: 8.5 -.2 -.3 -.3 -.4 -.5 -.4 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 00 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 00 Consumer prices, difference in growth rates Real consumer wage, difference in growth rates .025 .002 .020 .000 -.002 .015 -.004 .010 -.006 .005 -.008 .000 -.010 -.005 -.012 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 00 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 00 Figure 16: Economic responses due to floods under RCP 2.6 and RCP 8.5 6 Concluding remarks This paper represents the extension of a macroeconomic model to capture the impacts of natural disasters. It includes several changes that are essential to analyze the im- pact of natural disasters on economic activity, taking into account the specific of disaster- caused capital damages: (1) The capital stock is disaggregated into infrastructure and 28 non-infrastructure capital. (2) Private and public sector investment decisions into recon- struction are separated and explicitly modeled. (3) The impact of the shock on the produc- tivity of non-affected capital is explicitly introduced. (4) Realistic constraints on the pace of reconstruction are taken into account. The marginal product of capital plays an important role in investment allocation de- cisions. The productivity of infrastructure capital increases significantly after a natural disaster, while the implications for non-infrastructure capital depend crucially on the elas- ticity of substitution between types of capital. Empirical estimates suggest that the two capital stocks are complementary. Thus, if a natural disaster destroys infrastructure capi- tal, then non-infrastructure capital also becomes unproductive, which magnifies the impact of the disaster on GDP. In a simpler model where the capital stock is represented with a unique variable K , this effect would be equivalent to a loss in capital and a loss of total factor productivity (which is consistent with empirical observations). In the model, large-scale disasters, exceeding the absorptive capacity of the economy, generate total (discounted) consumption losses that respond non-linearly to (and may ex- ceed) direct physical damage. This result is also consistent with empirical evidence. For small disasters, the economic system can absorb the impact thanks to resource realloca- tion, imports, and borrowing, such that the total welfare cost is similar to or even lower than the direct physical damages.10 For larger disasters, the economy reaches the limits of its ability to reallocate resources, the reconstruction period extends over several years, and direct physical impacts are magnified. The absorptive capacity of the economy is controlled by a few parameters, including the ability of the economy to reallocate resources toward reconstruction and to implement the appropriate monetary policy response. The destruction of capital (and the increase in the marginal product of capital) gen- erates inflation, which can be more or less persistent depending on price stickiness. If monetary policy reacts immediately to the supply shock by increasing interest rates, eco- nomic effects are amplified. Monetary policy tightening increases the cost of borrowing for both the government and the private sector and reduces consumption. A delayed response (usually when second round effects materialize due to changes in demand) can mitigate the negative economic effects of disasters. However, in the context of climate change, increasingly frequent shocks may unanchor inflation expectations and reverse this result. Fiscal policy adds another layer of complexity. The impact of disasters has both quantity and price responses. In our simulations, the short-run tax base response is dominated by prices, which lead to an increase in the nominal tax base and, in some of the simulations, a temporary reduction in nominal debt. This revenue gain is only temporary, as the medium- to long-term response to the disaster shock involves a reduction in the tax base. It is beneficial, however, to invest in rebuilding the damaged capital, given that the return on this investment is particularly high (because it increases the productivity of the non- infrastructure assets). 10 This is ignoring the direct human impact of disasters, or assuming it can be avoided thanks to well- managed early warning systems and evacuation, or buildings designed not to collapse during an earthquake. 29 7 Appendix 7.1 Balanced growth path The balanced growth path of the economy depends on our population and total factor productivity assumptions, both which are exogenous in this model. The projection period starts in 2021 using available and up to date data as of the time of writing this paper. A critical outcome for balanced growth path is that all real (in this case at constant prices) variables growth at the same rate. Figure 17 illustrates that the model reaches steady state by 2100. Balanced growth path % change 12 8 4 0 GDP -4 Potential GDP HH Cons. -8 Gov Cons. GFCF -12 Exports Imports -16 2025 2050 2075 2100 2125 2150 2175 2200 Figure 17: Balanced Growth: Baseline 7.2 Proof of the lemma Lemma .1. This framework is consistent with the intuitions of Hallegatte and Vogt-Schilb (2019) as reparation will always be preferred to building new capital. Proof. The marginal productivity of repairing is ∂Yt∗ ∂Yt∗ ∂KLt−1 Yt ∂KLt−1 − p =− p = ∂DSt−1 ∂KLt−1 ∂DSt−1 KS,t−1 ∂DStP −1 and the marginal productivity of adding new capital is ∂Yt∗ ∂Yt Yt∗ Yt ∂KLt−1 Yt∗ ∂KS,t−1 P = P + − . ∂KS,t−1 ∂KS,t− 1 Yt KS,t−1 ∂DStP −1 P Yt KS,t−1 30 One can note that ∂Yt∗ ∂Yt∗ Yt∗ ∂Yt ∂KS,t−1 P = − p + P − P ∂KS,t−1 ∂DSt−1 Yt ∂KS,t−1 KS,t−1 where   ∂Yt ∂KS,t−1 ∂KS,t−1  ∂Yt  − = −1   P P P ∂KS,t−1 KS,t ∂KS,t ∂KS,t−1  −1 −1   >0 <1 because of decreasing return of the Cobb-Douglas in K <0 Hence, ∂Yt∗ ∂Yt∗ < − . P ∂KS,t−1 ∂DStp−1 In words, the marginal productivity of repairing is always higher than the marginal pro- ductivity of adding new capital stocks. References Aiyagari, S. R. (1994). Uninsured idiosyncratic risk and aggregate saving. The Quarterly Journal of Economics, 109(3):659–684. Akao, K.-I. and Sakamoto, H. (2018). A theory of disasters and long-run growth. Journal of Economic Dynamics and Control, 95(C):89–109. An, Z., Kangur, A., and Papageorgiou, C. (2019). On the substitution of private and public capital in production. European Economic Review, 118:296–311. Bakkensen, L. and Barrage, L. (2021). Climate shocks, cyclones, and economic growth: Bridging the micro-macro gap. NBER Working Paper w24893, National Bureau of Eco- nomic Research. Baqaee, D. and Farhi, E. (2020). Nonlinear production networks with an application to the covid-19 crisis. Technical report, National Bureau of Economic Research. Barrot, J.-N. and Sauvagnat, J. (2016). Input specificity and the propagation of idiosyn- cratic shocks in production networks. The Quarterly Journal of Economics, 131(3):1543– 1592. Batten, S., Sowerbutts, R., and Tanaka, M. (2016). Let’s talk about the weather: the impact of climate change on central banks. Bewley, T. (1977). The permanent income hypothesis: A theoretical formulation. Journal of Economic Theory, 16(2):252–292. 31 Boehm, C. E., Flaaen, A., and Pandalai-Nayar, N. (2019). Input Linkages and the Transmis- sion of Shocks: Firm-Level Evidence from the 2011 T¯ ohoku Earthquake. The Review of Economics and Statistics, 101(1):60–75. Burke, M., Hsiang, S. M., and Miguel, E. (2015). Global non-linear effect of temperature on economic production. Nature, 527(7577):235–239. Burns, A., Campagne, B., Jooste, C., Stephan, D., and Bui, T. T. (2019). The world bank macro-fiscal model technical description. Policy Research Working Paper 8965, The World Bank, Washington, DC. Burns, A. and Jooste, C. (2019). Estimating and calibrating mfmod : A panel data approach to identifying the parameters of data poor countries in the world bank’s structural macro model. Policy Research Working Paper 8939, The World Bank, Washington, DC. Burns, A., Jooste, C., and Schwerhoff, G. (2020). Macroeconomic modeling of managing hurricane damage in the caribbean: The case of jamaica. Policy Research Working Paper 9505, The World Bank, Washington, DC. Camacho, C. and Sun, Y. (2017). Longterm decision making under the threat of earth- quakes. PSE Working Papers halshs-01670507, HAL. Cavallo, A., Cavallo, E., and Rigobon, R. (2014). Prices and supply disruptions during natural disasters. Review of Income and Wealth, 60:S449–S471. Colon, C., Hallegatte, S., and Rozenberg, J. (2021). Criticality analysis of a country’s trans- port network via an agent-based supply chain model. Nature Sustainability, 4(3):209– 215. Dell, M., Jones, B. F., and Olken, B. A. (2012). Temperature shocks and economic growth: Evidence from the last half century. American Economic Journal: Macroeco- nomics, 4(3):66–95. Dieppe, A., Kilic Celik, S., and Okou, C. (2020). Implications of major adverse events on productivity. Working Paper 9411, The World Bank. Dietz, S. and Stern, N. (2015). Endogenous growth, convexity of damage and climate risk: how nordhaus’ framework supports deep cuts in carbon emissions. The Economic Journal, 125(583):574–620. Doyle, L. and Noy, I. (2015). The short-run nationwide macroeconomic effects of the canterbury earthquakes. New Zealand Economic Papers, 49(2):134–156. Ehrmann, M. and Smets, F. (2003). Uncertain potential output: implications for monetary policy. Journal of Economic Dynamics and Control, 27(9):1611–1638. uller, C., Deryng, D., Chryssanthacopoulos, J., Boote, K. J., B¨ Elliott, J., M¨ uchner, M., Fos- ter, I., Glotter, M., Heinke, J., Iizumi, T., Izaurralde, R. C., Mueller, N. D., Ray, D. K., Rosenzweig, C., Ruane, A. C., and Sheffield, J. (2015). The global gridded crop model intercomparison: Data and modeling protocols for phase 1 (v1.0). Geosci. Model Dev., 8:261–277. 32 Fankhauser, S. and Tol, R. S. (2005). On climate change and economic growth. Resource and Energy Economics, 27(1):1–17. Feenstra, R. C., Inklaar, R., and Timmer, M. P. (2015). The next generation of the penn world table. American Economic Review, 105(10):3150–3182. Felbermayr, G. and Gr¨ oschl, J. (2014). Naturally negative: The growth effects of natural disasters. Journal of Development Economics, 111(C):92–106. Ferrero, G., Pietrunti, M., and Tiseno, A. (2019). Benefits of gradualism or costs of inaction? monetary policy in times of uncertainty. Monetary Policy in Times of Uncertainty (February 4, 2019). Bank of Italy Temi di Discussione (Working Paper) No, 1205. Ginn, W. (2021). Climate disasters and the macroeconomy: Does state-dependence mat- ter? evidence for the us. EconDisCliCha. Hallegatte, S. and Ghil, M. (2008). Natural disasters impacting a macroeconomic model with endogenous dynamics. Ecological Economics, 68(1):582–592. Hallegatte, S., Ghil, M., Dumas, P., and Hourcade, J.-C. (2008). Business cycles, bifurca- tions and chaos in a neo-classical model with investment dynamics. Journal of Economic Behavior & Organization, 67(1):57–77. Hallegatte, S., Hourcade, J.-C., and Dumas, P. (2007). Why economic dynamics matter in assessing climate change damages: Illustration on extreme events. Ecological Economics, 62(2):330–340. Hallegatte, S. and Vogt-Schilb, A. (2019). Are losses from natural disasters more than just asset losses? In Advances in Spatial and Economic Modeling of Disaster Impacts, pages 15–42. Springer. Hallegatte, S., Vogt-Schilb, A., Bangalore, M., and Rozenberg, J. (2017). Unbreakable : Building the Resilience of the Poor in the Face of Natural Disasters. Washington, dc: World bank, World Bank. Henriet, F., Hallegatte, S., and Tabourier, L. (2012). Firm-network characteristics and economic robustness to natural disasters. 36(1):150–167. Howell, W., Evans, P., Devereux, S., Sage, D., Smith, J., and Haegert, D. (1993). Absence of strong hla-dr/dq-dp linkage disequilibrium in the british and french canadian caucasoid populations. International Journal of Immunogenetics, 20(5):363–371. Hsiang, S. M. and Jina, A. S. (2014). The causal effect of environmental catastrophe on long-run economic growth: Evidence from 6,700 cyclones. Working Paper 20352, National Bureau of Economic Research. Ikefuji, M. and Horii, R. (2012). Natural disasters in a two-sector model of endogenous growth. Journal of Public Economics, 96(9-10):784–796. Inoue, H. and Todo, Y. (2019). Firm-level propagation of shocks through supply-chain networks. Nature Sustainability, 2(9):841–847. 33 IPCC (2012). Managing the risks of extreme events and disasters to advance climate change adaptation. a special report of working groups i and ii of the intergovernmental panel on climate change [field, c.b., v. barros, t.f. stocker, d. qin, d.j. dokken, k.l. ebi, m.d. mastrandrea, k.j. mach, g.-k. plattner, s.k. allen, m. tignor, and p.m. midgley (eds.)]. Technical report, IPCC, United Kingdom and New York, NY, USA. Isore, M. and Szczerbowicz, U. (2017). Disaster risk preference shifts in a new keyenesian model. Journal of Economic Dynamics and Control, 79:97–125. Klomp, J. (2020). Do natural disasters affect monetary policy? a quasi-experiment of earthquakes. Journal of Macroeconomics, 64:103164. Krebs, T. (2003a). Growth and Welfare Effects of Business Cycles in Economies with Id- iosyncratic Human Capital Risk. Review of Economic Dynamics, 6(4):846–868. Krebs, T. (2003b). Human capital risk and economic growth. The Quarterly Journal of Economics, 118(2):709–744. Lackner, S. (2018). Earthquakes and economic growth. FIW Working Paper 190, FIW - Research Centre International Economics. Marto, R., Papageorgiou, C., and Klyuev, V. (2018). Building resilience to natural disasters: An application to small developing states. Journal of Development Economics, 135:574– 586. Moore, F. C. and Diaz, D. B. (2015). Temperature impacts on economic growth warrant stringent mitigation policy. Nature Climate Change, 5(2):127–131. Myhre, G., Alterskjær, K., Stjern, C., Hodnebrog, , Marelle, L., Samset, B., Sillmann, J., Schaller, N., Fischer, E., Schulz, M., and Stohl, A. (2019). Frequency of extreme precipi- tation increases extensively with event rareness under global warming. Scientific Reports, 16063. uller-F¨ M¨ urstenberger, G. and Schumacher, I. (2015). Insurance and climate-driven extreme events. Journal of Economic Dynamics and Control, 54(C):59–73. Nordhaus, W. D. (1992). An optimal transition path for slowing climate change. Science, 20:1315–1319. Nordhaus, W. D. and Moffat, A. (2017). A Survey of Global Impacts of Climate Change: Replication, Survey Methods, and a Statistical Analysis. NBER Working Papers 23646, National Bureau of Economic Research, Inc. ¨ Ocal, A. (2019). Natural disasters in turkey: Social and economic perspective. International Journal of Disaster Risk Management, 1(1):51–61. Parker, M. (2018). The impact of disasters on inflation. Economics of Disasters and Climate Change, 2(1):21–48. Raddatz, C. (2007). Are external shocks responsible for the instability of output in low- income countries? Journal of Development Economics, 84(1):155–187. 34 Rao, A., Calder´on, A., Silva, V., Martins, L., and Paul, N. (2022). Earthquake risk assess- ment and retrofit scenarios for turkey, exposure model development probabilistic seismic risk analysis retrofit intervention scenarios. Technical report, Global Earthquake Model. Rose, A. and Liao, S.-Y. (2005). Modeling regional economic resilience to disasters: A computable general equilibrium analysis of water service disruptions. Journal of Regional Science, 45(1):75–112. Rose, A. and Wei, D. (2013). Estimating the economic consequences of a port shutdown: the special role of resilience. Economic Systems Research, 25(2):212–232. Rudebusch, G. D. et al. (2019). Climate change and the federal reserve. FRBSF Economic Letter, 9. Strobl, E. (2011). The economic growth impact of hurricanes: Evidence from u.s. coastal counties. The Review of Economics and Statistics, 93(2):575–589. UNISDR (2015). Global assessment report on disaster risk reduction. Last accessed 18 December 2021. Wood, A., Noy, I., and Parker, M. (2016). The canterbury rebuild five years on from the christchurch earthquake. Bulletin 3, Reserve Bank of New Zealand. World Bank (1999). Turkey marmara earthquake assessment. Report 27380, Turkey Coun- try Office. World Bank (2021). Poverty and macro economic impacts of climate shocks. Technical report, The World Bank. Wright, A. and Borda, P. (2016). Macroeconomic fluctuations under natural disaster shocks in central america and the caribbean. IDB Working Paper Series 463, IDB. 35