Policy Research Working Paper 10298 Natural Disasters and Fiscal Drought Lazar Milivojevic South Asia Region Office of the Chief Economist February 2023 Policy Research Working Paper 10298 Abstract This paper examines to what extent slowdowns in economic A dynamic stochastic general equilibrium model is used to growth after natural disasters are accompanied by widening show the propagation mechanism of an extreme event that fiscal deficits and corresponding pressures on public debt. affects agricultural productivity. The model features farmers Empirical analysis based on exogenous measures of physical endowed with land with time-varying productivity subject disaster intensity shows that natural disasters lead not only to economic and weather conditions. Simulation results to output losses but also to further deterioration of countries’ illustrate the climate-fiscal nexus existence and highlight fiscal positions. The effects are persistent and driven by devel- the role of structural resilience in limiting the impact of opments in emerging markets and developing economies. natural disasters. This paper is a product of the Office of the Chief Economist, South Asia Region. 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 author may be contacted at lmilivojevic@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 Natural Disasters and Fiscal Drought Lazar Milivojevic∗ The World Bank Keywords: natural disasters, fiscal sustainability, climate change JEL-Codes: C51, E62, Q54. ∗ The author thanks Robert Beyer, Alexander Meyer-Gohde, Gauthier Vermandel, Volker Wieland, and participants at the Asian Economic Development Conference (ADB), 85th Anniversary Conference at the Faculty of Economics and Business (University of Belgrade), and South Asia Chief Economist Office internal seminar (World Bank) for helpful comments. 1 Introduction The COVID-19 pandemic has raised the awareness that international cooperation is indis- pensable to addressing urgent global challenges. Climate change is one of them. The fre- quency and severity of climate shocks, such as tropical cyclones, droughts, heat waves, or flooding, have intensified and are projected to worsen as a result of gradual global warm- ing (Stern, 2007; IMF, 2017; IPCC, 2018). The developing countries, and particularly their poorest segments of society, are likely to bear the worst impacts of climate-related catastrophes (Hallegatte et al., 2016). Climate change-related risks can be grouped into two categories: physical risks asso- ciated with changes in climate, and transition risks from the transition to a low-carbon economy (Feyen et al., 2020). Physical impacts, which are the focus of this paper, may substantially damage medium-term growth potential and contribute to public debt in- crease (IMF, 2017). This year’s heatwave and devastating floods in Pakistan have caused more than 1,700 deaths and displaced more than 8 million people. Economic losses and damage to infrastructure, assets, crops, and livestock have been worth more than 30 USD billion. Developing economies are particularly vulnerable due to their structural charac- teristics - less sectoral diversification and high informality level. The relative size of the agriculture sector and more households that are not protected against transitory reduc- tion in income could make the situation dire. Reduced economic activity might shrink fiscal revenues, while post-disaster relief and reconstruction spending require a rise in public expenditures, adding additional uncertainty to fiscal management. Against that backdrop, fiscal resilience is becoming even more important. Countries without fiscal space are constrained in their reactive capacity. Moreover, fiscal vulnerabilities reduce their scope for climate mitigation and adaptation policies. Figure 1: Climate-fiscal nexus Notes: The sample includes 128 countries in 2020, with higher values representing higher risk. AEs stands for advanced economies and EMDEs for emerging market and developing economies. The trendline is based on the whole sample. Source: IMF Climate Change Dashboard, EIU Country Risk Model, and Feyen et al. (2020). 1 To illustrate these challenges, Figure 1 shows that countries with higher vulnerabil- ity to climate change also face higher sovereign risks. The climate vulnerability index comes from the new IMF Climate Change Dashboard and comprises climate-driven haz- ard and exposure, vulnerability, and lack of coping capacity. The EIU Country Risk Index indicates sovereign debt risks, based on more than 50 institutional, macrofinancial and structural variables. Climate-fiscal nexus due to the simultaneous exposure to both risks suggests that emerging market and developing economies (EMDEs) are more sensitive to physical risks of climate change and feature tighter fiscal space. When natural disasters hit an economy, a fiscal response to address economic and distributional effects may lead to a debt increase which carries downside risks of its own. Potential debt distress and limited access to international financial markets could reduce the intended fiscal support and exacerbate growth and fiscal prospects. This paper examines to what extent the impacts on economic growth after natural disasters are accompanied by widening fiscal deficits and corresponding pressures on pub- lic debt. It relies on the ifo GAME database (Felbermayr and Groeschl, 2014), which contains exogenous measures of physical disaster intensity, well suited for causal analysis. Regression results suggest that natural disasters lead not only to output losses, but also to further deterioration of countries’ fiscal positions. A disaster of one standard deviation above the disaster index mean leads to 1.32 percent lower output per capita growth, 1.58 percent higher government debt, and 0.88 percent increase in fiscal deficit. Local projec- tion estimates indicate that the adverse effects persist in the years after the disaster, not only on impact. Finally, the effects are significant in the subsample of EMDEs, while the advanced economies remain stable after disaster episodes. In addition, I provide a general equilibrium interpretation of the empirical results using a Dynamic stochastic general equilibrium (DSGE) model. The model explains the propagation mechanism of a natural disaster shock that affects agricultural productivity. Based on Gallic and Vermandel (2020), it features farmers endowed with land with time- varying productivity subject to economic and weather conditions. The model is extended with a more detailed government sector, which allows considering the fiscal implications. Last but not least, counterfactual simulations highlight the role of structural resilience in limiting the impact of disasters. 2 Related Literature Empirical literature remains inconclusive on the macro-fiscal effects of natural disasters. When it comes to the impact on economic activity, Rasmussen (2004) suggests that natural disasters cause a 2.2 percent median reduction of the same-year real GDP growth rate. In a similar vein, Loayza et al. (2012) and Fomby et al. (2013) use a large panel of countries to conclude that severe disasters carry much worse effects than moderate ones, especially in developing economies. Klomp and Valckx (2014) conduct a meta-analysis based on 25 2 primary studies and find that natural disasters have a significant negative effect on growth per capita, particularly strong for climatic disasters in developing countries. Felbermayr and Groeschl (2014) propose a new panel dataset and find robust adverse nonlinear effects on economic activity. On the other hand, Skidmore and Toya (2002) and Cuaresma et al. (2008) find an expansionary effect attributable to a productivity boost from new investments and technologies. Research on how climate change affects the fiscal sector is relatively scarce. Most studies draw on the publicly available cross-country database on disaster losses called Emergency Events Database (EM-DAT). Lis and Nickel (2009) explore the implications of climate change for fiscal policy by assessing the impact of large-scale extreme weather events on changes in public budgets and conclude that the budgetary impact of such events ranges between 0.23 and 1.1 percent of GDP depending on the country group. Melecky and Raddatz (2011) examine the economic and fiscal consequences of a natural disaster for middle and high-income countries over the period 1975 through 2008. They divide the disasters into geological, climate, and a residual category, and find that on average budget deficits increase only after climate disasters, whereas for lower-middle-income countries, the increase in deficits is widespread across all events. Gerling (2017), however, finds that after weather-related disasters government primary balances remain surprisingly stable in a sample of 19 countries in Developing Asia from 1970 to 2015. Koetsier (2017) applies a panel synthetic control method to compare the difference in government debt between the disaster countries and their respective control groups. His dataset comprises 163 countries for the period between 1971 and 2014 and finds a considerable increase in government debt for the most damaging and deadliest disasters. Government debt, on average, increases by 11.3 percent of GDP compared to the synthetic control group, with a median effect of 6.8 percent of GDP. Feyen et al. (2020) consider a large sample of countries and conclude that the debt of the general government is 2.4 percent of GDP higher one year after the disaster. Additionally, sovereign ratings are lower two years later for countries that faced extreme weather events, consistent with the significant rise of the public debt level after one year. Other studies look at the effects on alternative indicators of public finance soundness. Cevik and Jalles (2020a) find that vulnerability and resilience to climate change have a significant impact on the government borrowing cost. They show that countries with greater vulnerability to risks associated with climate change have higher bond yields and spreads. Additionally, they find that climate change vulnerability and resilience have significant effects on the probability of sovereign debt default (Cevik and Jalles, 2020b) and sovereign credit ratings. The adverse effect is amplified in the case of developing countries owing to their weaker capacity to adapt to the consequences of climate change (Cevik and Jalles, 2020c). Klomp (2017) empirical analysis also indicates that large-scale natural disasters considerably increase the onset probability of a sovereign debt default by about three percentage points, with particularly widespread damage by major earthquakes 3 and storms. Theoretical literature on the economic effects of natural disasters has been gradually evolving. Marto et al. (2018) develop a model that explores the macroeconomic im- pact of a major natural disaster. Their study focuses on debt sustainability concerns that arise from the need to fully rebuild public infrastructure over the medium term and compares policies, such as building adaptation infrastructure and fiscal buffers, with the post-disaster support provided by donors. Burns et al. (2021) develop a macrostructural model for Jamaica that embeds climate damages from natural disasters and considers different risk management strategies to address them. When it comes to DSGE models, Keen and Pakko (2011) develop a model to investigate the appropriate monetary policy response to natural disasters like Hurricane Katrina. Fern´ andez-Villaverde and Levintal (2018) compare different solution methods for computing the equilibrium of DSGE mod- els with rare disasters, given large nonlinearities that could be triggered by these events. Strulik and Trimborn (2019) examine the welfare effects in the aftermath of natural dis- asters. Similarly, Cantelmo et al. (2019) study the macroeconomic outcomes and welfare implications in disaster-prone countries and policies to mitigate them. Most of the pa- pers focus on the channel in which disasters destroy durable goods or productive capital. Gallic and Vermandel (2020), on the other hand, build and estimate a DSGE model with a weather-dependent agriculture sector to investigate the adverse weather implications in New Zealand. The contribution of this paper is twofold. First, it provides evidence of the fiscal im- plications of natural disasters in a broad sample of countries using different specifications and the disaster measure that is based on physical intensity. As pointed out in Felbermayr and Groeschl (2014), EM-DAT and similar databases comprise disaster intensity measures that are a function of economic development. Reported monetary damages of disasters are higher in richer economies, while human costs are usually lower. As a consequence, this is likely to lead to biased estimates. Second, the paper provides a theoretical explanation of the obtained results by looking at the propagation of the shock that affects agricul- tural productivity. I add Gallic and Vermandel’s (2020) agriculture sector framework to an otherwise standard Real Business Cycle (RBC) model (Kydland and Prescott, 1982; King et al., 1988; King and Rebelo, 1999), extend it with a more detailed fiscal sector, and calibrate the model to resemble the economy of Bangladesh - a country particularly sensitive to frequent weather-related natural disasters. 3 The Empirical Framework 3.1 Data The main variable in the analyses – the measure of disaster intensity, comes from the ifo GAME database (Felbermayr and Groeschl, 2014). It is based on primary geophysical 4 and meteorological sources, that comprise droughts, earthquakes, floods, storms, and temperature extremes in more than 100 countries for over 30 years.1 The disaster index exploits physical measures of disaster intensity: precipitation, Richter scale, wind speed, and temperature. As such, it is more suitable for estimating causal relations between natural disasters and economic variables. Other databases (EM-DAT, NatCatSERVICE, and similar) contain disaster indicators that are functions of economic development and, therefore, may lead to biased estimates (Felbermayr and Groeschl, 2014). More particularly, the weather variation represents the difference in monthly rainfall in mm, defined as the proportional deviation of total monthly rainfall from average monthly rainfall of the entire available period. Flooding events are measured by the positive difference in total monthly precipitation, while droughts are indicated with the dummy variable. It is equal to one if at least three subsequent months have rainfall below 50 percent of the long-run average monthly mean, or if at least five months within a year have rainfall below 50 percent of the long-run monthly mean, and zero otherwise. The database uses the maximum realization within a single earthquake event as the measure of the physical disaster intensity of that earthquake. Similarly, the maximum total wind speed in knots on a country basis is their disaster intensity measure for storms. Temperature extremes are measured as the percentage difference between the maximum temperature in one month from the corresponding long-run monthly mean. Strong positive deviations are interpreted as heat waves, while strong negative ones are defined as cold waves. The overall disaster index uses the inverse of the standard deviation of a disaster type within a country over all years as precision weights to assure that no single disaster component dominates the movement of the whole index. Furthermore, as the impact of a given disaster on the economy of a country clearly depends on the disaster intensity relative to the overall size of the country, the authors scale all respective disaster variables by land area, as suggested in Skidmore and Toya (2002). The rest of the data comes from the World Bank WDI. The exceptions are the Polity index which is taken from the Polity IV Project and gross government debt-to-GDP ratios from the IMF Historical Public Debt Database. The final dataset covers an unbalanced panel of 108 economies over the period between 1979 and 2010. 1 The unbalanced panel covers 108 countries over the period 1979-2010: Albania, Algeria, Angola, Argentina, Armenia, Australia, Azerbaijan, Bangladesh, Belarus, Belgium, Bolivia, Botswana, Brazil, Bulgaria, Burkina Faso, Burundi, Cameroon, Canada, Chad, China, Colombia, Congo, Rep., Costa Rica, ote d’Ivoire, Croatia, Cyprus, Czech Republic, Denmark, Dominican Republic, Ecuador, Egypt, Arab Cˆ Rep., El Salvador, Ethiopia, Finland, France, Gabon, Germany, Greece, Guatemala, Haiti, Honduras, Hungary, India, Indonesia, Ireland, Israel, Italy, Japan, Jordan, Kenya, Korea, Rep., Kuwait, Kyrgyz Republic, Lao PDR, Liberia, Madagascar, Malaysia, Mali, Mauritania, Mexico, Mongolia, Morocco, Mozambique, Namibia, Nepal, Netherlands, New Zealand, Nicaragua, Niger, North Macedonia, Norway, Oman, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Poland, Portugal, Romania, Russian Federation, Rwanda, Senegal, Serbia, Singapore, Slovak Republic, Slovenia, South Africa, Spain, Sri Lanka, Sweden, Switzerland, Syrian Arab Republic, Tajikistan, Tanzania, Thailand, Togo, Trinidad and Tobago, Tunisia, Uganda, Ukraine, United Kingdom, United States, Uruguay, Vietnam, Yemen, Rep. and Zambia. 5 3.2 Methodology The empirical analysis in this section uses fixed effect panel regressions and the local projection method. The aim is to estimate the average outcome of natural disasters on macro-fiscal variables, but also the effects in the medium run. I estimate the benchmark panel regressions first, following Felbermayr and Groeschl (2014): yi,t = ci + τt + βDi,t + θXi,t−1 + εi,t (1) where y denotes the dependent variable of interest – GDP per capita growth, change in government debt-to-GDP or fiscal balance. To ensure their stationarity, fiscal variables are expressed as a share of GDP and first-differenced.2 ci and τt represent country and time fixed effects, while D is a measure of disaster intensity. X stands for a vector of lagged variables that control for the structural, economic policy, and external factors (Skidmore and Toya, 2002; Noy, 2009; Loayza et al., 2012; Felbermayr and Groeschl, 2014). Control variables include the level of development (lagged GDP per capita), the size of the economy (total population), an index of democratization (polity index), openness to trade, inflation, real interest rates, domestic credit, gross capital formation, current account balance, and foreign direct investment. I also consider two regressions with budgetary components as dependent variables - changes in total revenues and expenditures as a share of GDP are also considered. Finally, separate regressions are estimated for the subsamples of advanced economies and EMDEs. In all cases, country-clustered robust standard errors are used for statistical inference. In addition, I run the local projection method (Jord` a, 2005) to estimate the dynamic effect in the years after the shock. This method provides several advantages over the tra- ditional structural VAR methodology. It estimates sequential regressions of the dependent variable shifted several steps ahead instead of recursive use of the initial set of estimated coefficients. It is, therefore, more robust to potential misspecifications (Auerbach and Gorodnichenko, 2013; Romer and Romer, 2015). The specification is the following: yi,t+k − yi,t−1 = ci + τt + βk Di,t + θXi,t±l + εi,t (2) where k=0,..,4, that is, up to four years after the disaster. The same notation and the set of variables are used as before, while βk indicates the cumulative response of y in each k year after disasters. Additionally, to correct for the potential bias described in Teulings and Zubanov (2014), I include the measure of disaster intensity for each period between t+1 and t+k. Finally, to compute confidence intervals associated with the estimated coefficients, country-clustered robust standard errors are used again. 2 Standard F-tests reject the null hypothesis of a unit root presence in considered series, as well as Levin–Lin–Chu unit root tests on balanced subsamples. 6 3.3 Results Natural disasters lead to an overall decrease in economic activity. Additionally, they cause statistically significant fiscal deterioration – an increase in both government debt and fiscal deficit as a share of GDP. Table 1: Benchmark regression results Notes: Table shows the outcome of three estimated fixed effect panel regressions. Source: Author’s calculations. Table 1 shows that natural disasters have negative impact on per capita output growth, fiscal balance, and government debt-to-GDP. More specifically, the mean and standard deviation of the disaster index leads to the following interpretation: a disaster equal to its sample mean (0.031) decreases output per capita growth by 0.21 percent on average, while the adverse effect on debt and fiscal balance (both as a share of GDP) is 0.25 and -0.14 percent, respectively. A disaster of one standard deviation (0.17) above the mean has more severe consequences – 1.32 percent lower output per capita growth, 1.58 percent 7 higher government debt, and 0.88 percent increase in fiscal deficit. To disentangle the impact on fiscal balance, I estimate two additional regressions that consider its components - total revenues and expenditures as a share of GDP. Table 2 indicates that government total revenues remain stable after disaster episodes. This may reflect several factors (see Gerling (2017) for more details). For example, the disaster may hit parts of the economic activity that are not subject to tax payments anyway (informal sector or subsistence farmers). Besides, more ambitious revenue mobilization efforts on the authorities’ side could offset a decrease in the tax base. Additional budget resources could also come in the form of external official grants. On the other hand, the change in total expenditures is statistically significant. This result could be explained by the post-disaster relief measures and the government motivation to address adverse aggregate demand and distributional impacts. Table 2: Regression results for the budget components Notes: Table shows the outcome of estimated fixed effect panel regressions for total revenues and total expenditures as a share of GDP. Source: Author’s calculations. In addition, it is possible to estimate the effects in the two subsamples of countries - advanced economies and EMDEs. Although disasters affect macro-fiscal variables in the same direction as in the benchmark regressions, none of the effects turns out to be statistically significant in advanced economies.3 On the contrary, Table 3 indicates significant impacts in EMDEs. This is not surprising, given the structural characteristics of their economies, vulnerability to disasters, and lack of capacity to cope with them. A disaster of one standard deviation (0.19) above the mean (0.037) in EMDEs would result in 1.58 percent lower output per capita growth, 2.1 percent higher government debt, and 0.68 percent worsening in fiscal balance. The contemporaneous effect of disasters is well-documented in the literature, given the immediate damage to the capital stock and the productivity of assets such as land. However, the persistence of that effect has been questioned. Local projection estimation results imply that the consequences remain severe in the years after disasters (Figure 2). One standard deviation above the mean disaster yields a contemporaneous output per capita growth decline of 1.19 percent and a cumulative decrease of 2.87 percent in the 3 The results are available upon request. 8 Table 3: Benchmark regression results for EMDEs Notes: Table shows the outcome of three estimated fixed effect panel regressions in the sample of EMDEs. Source: Author’s calculations. medium term. Government debt-to-GDP ratio increases by 1.66 percent on impact and reaches 3.28 percent after three years. Finally, fiscal balance also increases by around 1 percent of GDP. These persistent economic implications could have their roots in higher productivity of destroyed capital (Hallegatte and Vogt-Schilb, 2019) or in a crowding- out of productive capital expenditures by reconstruction efforts (Raddatz, 2007; Noy, 2009). Fiscal deterioration may come from relief and reconstruction efforts financing and a potential worsening of borrowing conditions (Klomp, 2017). Explaining the drivers of the obtained dynamics is beyond the scope of this paper, however. Figure 2: Cumulative impulse responses to a one standard deviation increase of the index Notes: Figure shows impulse responses and 90-percent confidence intervals from the estimated local projection regressions. Source: Author’s calculations. While certain disasters affect larger portions of the economy, some of them, like droughts, floods, or extreme weather events, are particularly damaging for the agriculture sector. The estimated interaction coefficients in Appendix indeed indicate that the higher reliance on agriculture, expressed either as a share to GDP or employment share, has stronger negative output outcomes. That channel is further explored from the theoretical perspective in the next section. When it comes to the robustness checks, the regressions are re-estimated using the smaller set of explanatory variables or different lag structures. None of those alters the previous conclusions. I do not explicitly deal with the bias that arises from the presence 9 of a lagged endogenous variable on the right-hand-side of the equations (Nickell, 1981), since this bias is negligible in panels with longer time dimensions (Judson and Owen, 1999; Felbermayr and Groeschl, 2014). 4 The Structural Approach This section proposes a DSGE model that explains the transmission mechanism of disas- ters that affect the agriculture sector. One could think of a shock in the form of adverse weather events. I rely on the standard RBC model that features weather-dependent agricultural productivity and exogenous weather shocks, based on Gallic and Vermandel (2020). The model describes a two-sector (agricultural and non-agricultural), two-good economy, populated by households and firms. Crucially, workers from the agriculture sec- tor face unexpected weather events that affect the productivity of their land. Finally, the government sets the countercyclical fiscal policy. The remainder of the section provides an overview of the model’s structure, its calibration, and proceeds with the simulations and comparison exercises. 4.1 The model Households There is a continuum j ∈ [0, 1] of identical households that maximize their lifetime utility. They consume, save, and work in the two production sectors. The lifetime util- ity function is separable in consumption and individually supplied labor and takes the following form: ∞ 1 χ Et βτ (Cj,t+τ − bCt+τ −1 )1−σ − h1+σH (3) τ =0 1−σ 1 + σH j,t+τ where the variables C and h represent the consumption index and the labor effort index. β denotes the discount factor, σ the relative risk aversion, σH is the labor disutility coefficient, while b accounts for external consumption habits. Labor supply is affected by a shift parameter χ that pins down the steady state of hours worked. As in Gallic and Vermandel (2020) and following Horvath (2000), there is a CES labor disutility index with imperfect substitutability of labor supply between the agriculture and non-agriculture sectors: 1 1+ι 1+ι (1+ι) hj,t = hN j,t + hA j,t (4) The labor disutility index consists of hours worked in the non-agriculture and agri- culture sectors, with a costly relocating across them governed by the substitutability parameter ι. 10 The representative household faces the real budget constraint at every period t: s s wt hj,t + rt−1 Bj,t−1 − Tt = Cj,t + Bj,t (5) s=N,A where the income of the representative household comes from the provided labor ser- vice hs with a wage ws (in each sector s), and real risk-free domestic bonds B remunerated at a rate r. The household’s expenditures go to its consumption basket, bonds, and the lump-sum taxes T charged by the government. The representative household allocates total consumption between non-agricultural and agricultural consumption goods. The CES consumption bundle reads as follows: µ 1 µ−1 1 µ−1 (µ−1) N A Cj,t = (1 − ϕ) µ Cj,t µ + (ϕ) µ Cj,t µ (6) where µ denotes the substitution elasticity between the two corresponding types of goods, while ϕ is the fraction of agricultural goods in the total consumption basket. The consumption price index is determined accordingly by: µ 1 µ−1 1 µ−1 (µ−1) N A Pj,t = (1 − ϕ) µ Pj,t µ + (ϕ) µ Pj,t µ (7) Finally, demand for each type of good is a function of the total consumption index and its relative price: −µ −µ N A N PC,t A PC,t C j,t = (1 − ϕ) C j,t and C j,t =ϕ C j,t (8) Pt Pt Agriculture sector The economy is populated by a unit mass i ∈ [0, 1] of entrepreneurs, where a fraction n operate in the agricultural sector (farmers).4 The agricultural output is produced with the Cobb-Douglas production function: A ω α 1−α 1−ω yi,t = Ω εW t li,t−1 εZ A t ki,t−1 κA hA i,t (9) where y A is the outcome of farmers’ activity that combines an amount of land l , physical capital k A , and labor hA inputs. The production is subject to a weather shock through a function Ω εW t described in detail below and an economy-wide technology Z shock ε that follows an AR(1) process. The parameter ω is the elasticity of output to land, α stands for the physical capital share in the agricultural production, whereas κA is a technology parameter endogenously determined in the steady state. The farmers’ land l is subject to both economic and meteorological conditions. The agricultural production process between t-1 and t might be affected by the unexpected realization of the weather εW through a simple damage function: 4 Entrepreneurs could switch from one sector to another according to a stochastic AR(1) process. 11 −θ Ω εW t = εW t (10) where θ determines the elasticity of land productivity with respect to the weather, with lower values implying reduced propagation of weather-driven business cycles. To capture the implications of weather shocks as a source of aggregate fluctuations, the weather variable εW dynamics is introduced as: log εW W W t = ρW log εt−1 + σW ηt , W ηt ∼ N (0, 1) (11) where ρW and σW stand for the persistence of the weather shock and its standard deviation. In addition to the contemporaneous impact of weather events, agricultural production may be subject to effects that spread over time. For example, prolonged severe droughts entail early liquidation of stocks combined with a drop in the fertility rate (Kamber et al., 2013), while soil moisture deficits exhibit a persistence that is directly connected to the interaction between rainfalls and evapotranspiration, as lands require several months to recover their average productivity levels (Narasimhan and Srinivasan, 2005). Therefore, the productivity of land is assumed to be time-varying, following an endogenous law of motion: li,t = ((1 − δl ) + υ (xi,t )) li,t−1 Ω εW t (12) where δl is the rate of decay of land productivity that reflects the potential persistence effect. The marginal product of land is increasing in the accumulation of land productivity and captured by assuming that land expenditures x yield a gross output of new productive land υ (xi,t ) li,t−1 with υ > 0 and υ ≥ 0. These expenditures could be considered as agricultural spending on pesticides, herbicides, seeds, fertilizers, and water used to maintain farmland productivity. The functional form of land costs is given by: τ φ υ (xi,t ) = x (13) φ i,t where for φ → 0 land productivity exhibits constant returns, while for φ > 0 returns are increasing. The parameter τ allows to pin down the amount of per capita land in the steady state. The law of motion of agriculture sector physical capital has the usual form: A A A ki,t = (1 − δK ) ki,t −1 + ii,t (14) where δK is the depreciation rate of physical capital, whereas iA accounts for invest- ment. The profit maximization of the price-taking representative farmer can be cast as choos- ing the input levels under land efficiency, capital law of motion, and technology constraint: 12 ∞ max Et Λi,t+τ dA i,t+τ (15) { hA i,t , iA i,t , A , ki,t li,t ,xi,t } τ =0 where E denotes the expectation operator, Λ is the household stochastic discount factor between t and t+1, and d represents real profits that are given by: iA i,t dA A A N i,t = pt yi,t − pt iA i,t + S εi t iA i,t−1 A A − wt hi,t − pN t xi,t (16) iA i,t−1 in which pA is the relative production price of agricultural goods, whereas the function S (x) = 0.5κ (x − 1)2 denotes the convex investment cost function as in Christiano et al. (2005). Non-agriculture sector There is a continuum i of perfectly competitive non-agricultural producers with the following Cobb-Douglas production function5 : N α 1−α yi,t = εZ N t ki,t−1 hN i,t (17) where y N is the production of non-agricultural output in the ith firm, with the tech- nology εZ and input factors - physical capital k N and labor services hN . The parameters α and 1 − α denote output elasticity to capital and labor. Technology is assumed to be economy-wide (the same across sectors) and characterized as an AR(1) shock process: log εZ Z Z t = ρZ log εt−1 + σZ ηt (18) where ρZ and σZ represent the persistence and the standard deviation of the shock. The physical capital in the non-agricultural sector evolves in the following way: N N N ki,t = (1 − δK ) ki,t −1 + ii,t (19) where, as before, δK denotes the depreciation rate of physical capital, while iN is in- vestment from non-agricultural firms. The representative non-agricultural firm maximizes the discounted sum of profits, under capital accumulation and technology constraints: ∞ max Et Λi,t+τ dN i,t+τ (20) { hN i,t , iN i,t , N} ki,t τ =0 with real profits calculated as: 5 Non-agricultural firms do not require land inputs to produce goods and are not directly affected by weather. 13 iN i,t dN i,t = pN N t yi,t − pN t iN i,t +S εi t iN i,t−1 N N − wt hi,t (21) iN i,t−1 Fiscal authority The government spends on non-agricultural output and collects lump-sum taxes. It also pays interest on the previous stock of debt and issues new debt. At every period, there is the following budget constraint: Gt + rt−1 Bt−1 = Bt + Tt (22) The value of government spending follows the process described below: Gt − G Yt − Y = −φG (23) G Y assuming that the non-stochastic steady-state value (G) is an exogenous fraction of steady-state output. The government sets its policy in a countercyclical manner - when output is below its steady state, the government spending increases. Given that the weather shock is the only one considered in the model, this aims to resemble fiscal support to post-disaster recovery efforts, including rebuilding damaged infrastructure and provid- ing social assistance to affected households. The parameter φG measures the strength of the government response. Following Gali et al. (2007) and Cogan et al. (2010), the lump-sum tax reacts to respective deviations in the existing value of debt and government spending: Tt − T Bt − B Gt − G = ρB + ρG (24) Y Y Y where ρB and ρG are positive constants. The non-stochastic steady-state value of debt B ( Y ) is set to be in line with the observed data. 4.2 Aggregation and equilibrium In addition to the first-order conditions from the household and entrepreneurs’ optimiza- tion problems, market clearing conditions are necessary to set the standard equilibrium equations. The non-agricultural products market clearance is determined when the aggregate supply is equal to aggregate demand: −µ N PC,t (1 − nt ) YtN = (1 − ϕ) Ct + Gt + (1 − nt ) ItN + nt ItA + nt xt (25) Pt suggesting that the total supply of non-agricultural goods is equal to the respective household and government consumption, aggregate investment, and agricultural expen- 14 ditures. The equilibrium of the agricultural goods market, on the other hand, is given by: µ A PC,t nt YtA =ϕ Ct (26) Pt In sum, aggregate real production is determined as a combination of both: Yt = (1 − nt ) pN t Yt N + nt pA t Yt A (27) Given the presence of intermediate inputs, the GDP is calculated as: gdpt = Yt − pN t nt xt (28) The labor market clearing condition between household labor supply and demand in each sector enables writing the total number of hours worked as: Ht = (1 − nt ) HtN + nt HtA (29) 4.3 Calibration The model is log-linearized and calibrated in quarterly frequency to resemble the econ- omy of Bangladesh (Table 4). Some parameter values are set such that the steady-state conditions are consistent with long-run historical averages and microeconomic evidence. Other parameters are calibrated in line with the values commonly used in the literature for developing and developed countries or according to Gallic and Vermandel’s (2020) estimations. More specifically, β is set to 0.98, as Ahmad et al. (2012) find similar values for the long-run discount factor in a group of developed and developing economies. The depreciation rate of physical capital (δK ) is 0.025, with the capital share in technology (α) of 0.31, both as in Amin et al. (2018, 2019). Turning to the agriculture sector, the share of agricultural goods in the consumption basket of households (ϕ) is set to 50 percent, given the higher share of food products in developing economies’ consumption. In addition, the land-to-employment ratio is 0.12, based on the hectares of arable land per person in Bangladesh (World Bank WDI). Pa- rameter values imply the steady-state share of the agriculture sector close to the values observed in data (World Bank WDI). The elasticity of land productivity with respect to the weather (θ) is set to 20.59 with alternative values considered in the following subsec- tion. Finally, an extreme weather event is proxied by a weather shock with the persistence 0 and the standard deviation of 0.58, such that it results in the contemporaneous impact of 1 percent of GDP. When it comes to fiscal parameters, the steady-state share of spending in GDP is 0.14 (IMF WEO), while the debt-to-GDP ratio equals 30 percent (IMF WEO). Lump- sum tax function parameters (ρB and ρG ) are 0.33 and 0.1, as in Gali et al. (2007). 15 The government spending function reaction parameter (φG ) is set to 2. The comparison exercise that follows considers its alternative values. The rest of the values come from Gallic and Vermandel (2020). The risk aversion parameter (σ ) is 1.27, while the labor disutility coefficient (σH ) takes a value of 4.27. The steady-state share of hours worked per day is one-third, as usual in the literature. The consumption habits (b ) and the investment adjustment cost (κ) are equal to 0.59 and 2.44. The substitution cost across labor types (ι) is 2.85, whereas the substitution elasticity between two types of goods (µ) is 5.96. Additionally, the land cost function parameter (φ) is 3.57, the land efficiency decay rate (δl ) 0.05, and the share of land in the agricultural production function (ω ) is set to 0.3.6 Table 4: Baseline calibration 6 The share of land in the production function is estimated at 12 percent in New Zealand (Gallic and Vermandel, 2020), 15 percent in Canada (Echevarria, 1998), while Restuccia et al. (2008) calibrate it at 18 percent in the United States. The value here is set above those values. 16 4.4 Simulation results Figure 3 shows impulse responses from an unexpected extreme weather event and how it propagates through the economy. Essentially, the shock acts as a standard sectoral negative supply shock through a combination of rising hours worked and falling output. It results in a large decline in agricultural output, as land productivity is strongly affected. To offset this loss, farmers use more non-agricultural goods as inputs to reestablish their land productivity. This leads to an increase in non-agricultural output. As the weather shock causes a reduction in agricultural production and a rise in land costs, the relative price in the agriculture sector increases. The government responds to a decline in economic activity and increases its spending accordingly, which altogether leads to a surge in the debt-to-GDP ratio. Figure 3: Impulse responses to an extreme weather event Notes: Figure shows the model simulated impulse responses to an extreme weather event. Source: Author’s calculations. Structural resilience determines the degree to which disasters affect the economy. To proxy that effect, it is possible to compare the model’s implications for different elastic- ities of land productivity with respect to extreme weather events. The resulting impulse responses in Figure 4 suggest that both economic activity and fiscal stance are less af- fected when land productivity is more resilient, highlighting the need for infrastructure investments and adaptation policies in general. This supports the calls for structural mea- sures in sectors such as agriculture, water, and construction. In agriculture, this includes promoting farming practices that withstand climate variability, for example, the use of drought-resistant crop varieties or more efficient use of water resources, but also building up farmers’ capacity to adapt to long-term change (Freeman et al., 2003). 17 Figure 4: Impulse responses under different land sensitivity Notes: Figure is mostly illustrative, showing the model simulated impulse responses under different land sensitivity to weather conditions. Source: Author’s calculations. In addition, the model could be used to compare the alternative strengths of fiscal responses (Figure 5). The stronger the government impulse to the economy, reflected in the higher value of the parameter φG , the smaller the economic distress. However, the pressure on public finance could be elevated, emphasizing the need for the buildup of fiscal buffers in good times. In other words, the more fiscal space is available, the higher is the capacity of the government to mitigate the adverse effects of disaster episodes. Figure 5: Impulse responses under different fiscal reactions Notes: Figure is mostly illustrative, showing the model simulated impulse responses under different fiscal reaction parameters. Source: Author’s calculations. The simulations illustrate that risks associated with natural disasters may affect fis- cal sustainability in vulnerable countries. On the other hand, structural resilience and credible fiscal management should reduce these risks. The climate-fiscal nexus may be challenging in disaster-prone countries, raising the calls for international support to com- plement country-specific climate-related efforts. 18 5 Conclusion The analyses in this paper show that natural disasters lead to fiscal space shrinkage along with a slowdown in economic growth. A disaster of one standard deviation above the mean of the disaster index reduces output growth per capita by 1.32 percent and increases debt- to-GDP by 1.58 percentage points. The results are driven by developments in EMDEs and persist in the medium term. The observed climate-fiscal nexus implies that disaster- prone countries may feature a limited capacity to respond to extreme events, but also to build up the necessary resilience. Disaster relief and infrastructure rebuilding require fiscal costs that would add implicit charges on available space, emphasizing the need to supplement future fiscal assessments with the context of climate change (Heller, 2020). The model simulations illustrate the importance of building stronger resilience in coun- tries vulnerable to severe natural disasters. That requires a multipronged approach (IMF, 2019; Heller, 2020): first, better structural resilience by investing in both “hard” policy measures (e.g. physical infrastructure) and “soft” measures (such as establishing early warning systems); second, building financial resilience to protect fiscal sustainability and manage recovery costs, which includes self-insurance in the form of the provision of fiscal buffers and the transfer of risk through insurance and other risk-sharing mechanisms; and finally, creating post-disaster (and social) resilience that requires contingency planning to ensure a speedy response to a disaster. Climate change is inevitable, but better policy management should help to cope with its consequences for public finances in particular and economic development in general (Feyen et al., 2020). A comprehensive climate financing strategy will be necessary to adapt to the physical effects of climate change. Neither public money nor the private sector alone can do that job. Concessional funding, combining public and private capital through innovative financial instruments, and appropriate policies to align incentives with climate objectives must have their roles. International financial institutions will also have to step in through funding, technical assistance, and capacity development. There are several ways to extend the analyses in the paper. One is to consider al- ternative propagation channels of disaster impacts, for example, through the effects on productive capital. Another possibility could be to discuss different government man- agement strategies for disaster damages. Comparing adaptation investments, purchasing insurance from the private sector, or creating fiscal buffers in anticipation of responding to shocks, could provide valuable insights for policy makers. 19 References Acevedo, S., 2014.Debt, growth and natural disasters a Caribbean trilogy. IMF Work- ing Paper No. 14/125. International Monetary Fund. Acevedo, S., Mrkaic, M., Novta, N., Pugacheva, E. and Topalova, P., 2020. The Effects of Weather Shocks on Economic Activity: What are the Channels of Impact?. Journal of Macroeconomics, 65, p.103207. Ahmad, W., Haider, A., and Iqbal, J., 2012. Estimation of Discount Factor (beta) and Coefficient of Relative Risk Aversion (gamma) in Selected Countries, Munich Personnel RePEc Archive, MPRA Paper No. 39736. Amin, S., Marsiliani, L. and Renstr¨ om, T., 2018. The Impacts of Fossil Fuel Subsidy Removal on Bangladesh Economy. Bangladesh Development Studies, 41 (02), pp.65-81. Amin, S., Jamasb, T., Llorca, M., Marsiliani, L. and Renstr¨ om, T.I., 2019. Combining Private and Public Resources: Captive Power Plants and Electricity Sector Development in Bangladesh. Emerging Markets Finance and Trade, pp.1-22. Auerbach, A.J. and Gorodnichenko, Y., 2013. Output spillovers from fiscal pol- icy. American Economic Review, 103 (3), pp.141-46. Botzen, W.W., Deschenes, O. and Sanders, M., 2019. The economic impacts of natural disasters: A review of models and empirical studies. Review of Environmental Economics and Policy, 13 (2), pp.167-188. Burns, A, Jooste, C, and Schwerhoff, G., 2021. Macroeconomic Modeling of Managing Hurricane Damage in the Caribbean: The Case of Jamaica. Policy Research Working Paper No. 9505. World Bank, Washington, DC. Cantelmo, M.A., Melina, M.G. and Papageorgiou, M.C., 2019. Macroeconomic out- comes in disaster-prone countries. International Monetary Fund. Cevik, S. and Jalles, J.T., 2020. This changes everything: climate shocks and sovereign bonds. Cevik, S. and Jalles, J.T., 2020. Feeling the heat: Climate shocks and credit ratings. Cevik, S. and Jalles, J.T., 2020. An Apocalypse Foretold: Climate Shocks and Sovereign Defaults. Christiano, L.J., Eichenbaum, M. and Evans, C.L., 2005. Nominal rigidities and the dynamic effects of a shock to monetary policy. Journal of political Economy, 113 (1), pp.1-45. Cogan, J.F., Cwik, T., Taylor, J.B. and Wieland, V., 2010. New Keynesian ver- sus old Keynesian government spending multipliers. Journal of Economic Dynamics and Control, 34 (3), pp.281-295. Cuaresma, J.C., J. Hlouskova, and M. Obersteiner, 2008. Natural Disasters as Creative Destruction? Evidence from Developing Countries. Economic Inquiry, vol. 46, pp.214- 226. 20 Echevarria, C., 1998. A three-factor agricultural production function: the case of Canada. International Economic Journal, 12 (3), pp.63-75. Felbermayr, G. and Gr¨ oschl, J., 2014. Naturally negative: The growth effects of natural disasters. Journal of Development Economics, 111, pp.92-106. Fern´andez-Villaverde, J. and Levintal, O., 2018. Solution methods for models with rare disasters. Quantitative Economics, 9(2), pp.903-944. Feyen, E., Utz, R., Zuccardi Huertas, I., Bogdan, O. and Moon, J., 2020. Macro- financial aspects of climate change. The World Bank. Fomby, T., Ikeda, Y. and Loayza, N.V., 2013. The growth aftermath of natural disasters. Journal of applied econometrics, 28 (3), pp.412-434. Freeman, P.K., Keen, M., and Mani, M., 2003. Dealing with increased risk of natural disasters: challenges and options. Gal´ opez-Salido, J.D. and Vall´ ı, J., L´ es, J., 2007. Understanding the effects of gov- ernment spending on consumption. Journal of the European Economic Association, 5 (1), pp.227-270. Gallic, E. and Vermandel, G., 2020. Weather shocks. European Economic Review, 124, p.103409. Gerling, M.K., 2017. The Macro-Fiscal Aftermath of Weather-Related Disasters: Do Loss Dimensions Matter? International Monetary Fund. Hallegatte, S., Vogt-Schilb, A., Bangalore, M. and Rozenberg, J., 2016. Unbreak- able: building the resilience of the poor in the face of natural disasters. World Bank Publications. 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 (pp. 15-42). Springer, Cham. Heller, P.S., 2020. Scenarios for Fiscal Space: Adapting to the Long-Term Challenges of Climate Change. Horvath, M., 2000. Sectoral shocks and aggregate fluctuations. Journal of Monetary Economics, 45 (1), pp.69-106. International Monetary Fund. 2017. Seeking Sustainable Growth: Short-Term Recov- ery, Long-Term Challenges. Washington, DC, October. International Monetary Fund, 2019. Building Resilience in Developing Countries Vul- nerable to Large Natural Disasters, IMF Policy Paper (June). IPCC, 2018: Global warming of 1.5◦ C. An IPCC Special Report on the impacts of global warming of 1.5◦ C above pre-industrial levels and related global greenhouse gas emission pathways, in the context of strengthening the global response to the threat of climate change, sustainable development, and efforts to eradicate poverty. In Press. Jord` ` 2005. Estimation and inference of impulse responses by local projec- a, O., tions. American Economic Review, 95 (1), pp.161-182. Jord` ` and Taylor, A.M., 2016. The time for austerity: estimating the average a, O. 21 treatment effect of fiscal policy. The Economic Journal, 126 (590), pp.219-255. Judson, R.A. and Owen, A.L., 1999. Estimating dynamic panel data models: a guide for macroeconomists. Economics letters, 65 (1), pp.9-15. Kamber, G., McDonald, C., Price, G., et al., 2013. Drying out: Investigating the economic effects of drought in New Zealand. Reserve Bank of New Zealand Analytical Note Series No. AN2013/02. Keen, B.D. and Pakko, M.R., 2011. Monetary policy and natural disasters in a DSGE model. Southern Economic Journal, 77(4), pp.973-990. King, R.G., Plosser, C.I. and Rebelo, S.T., 1988. Production, growth and business cycles: I. The basic neoclassical model. Journal of monetary Economics, 21 (2-3), pp.195- 232. King, R.G. and Rebelo, S.T., 1999. Resuscitating real business cycles. Handbook of macroeconomics, 1, pp.927-1007. Kling, G., Lo, Y.C., Murinde, V. and Volz, U., 2018. Climate vulnerability and the cost of debt. Available at SSRN 3198093. Klomp, J. and Valckx, K., 2014. Natural disasters and economic growth: A meta- analysis. Global Environmental Change, 26, pp.183-195. Klomp, J., 2017. Flooded with debt. Journal of International Money and Finance, 73, pp.93-103. Koetsier, I., 2017. The fiscal impact of natural disasters. USE Discussion paper se- ries, 17 (17). Kydland, F.E. and Prescott, E.C., 1982. Time to build and aggregate fluctuations. Econo- metrica: Journal of the Econometric Society, pp.1345-1370. Lis, E.M. and Nickel, C., 2009. The impact of extreme weather events on budget balances and implications for fiscal policy. Loayza, N.V., Olaberria, E., Rigolini, J. and Christiaensen, L., 2012. Natural disasters and growth: Going beyond the averages. World Development, 40 (7), pp.1317-1336. Marto, R., Papageorgiou, C. and Klyuev, V., 2018. Building resilience to natural disas- ters: An application to small developing states. Journal of Development Economics, 135, pp.574-586. Melecky, M. and Raddatz, C., 2011. How do governments respond after catastrophes? Natural-disaster shocks and the fiscal stance. The World Bank. Narasimhan, B. and Srinivasan, R., 2005. Development and evaluation of Soil Mois- ture Deficit Index (SMDI) and Evapotranspiration Deficit Index (ETDI) for agricultural drought monitoring. Agricultural and forest meteorology, 133 (1-4), pp.69-88. Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica: Journal of the Econometric Society, pp.1417-1426. Noy, I., 2009. The Macroeconomic Consequences of Disasters. Journal of Development Economics, vol. 88, pp. 221-231. Noy, I. and Nualsri, A., 2011. Fiscal storms: public spending and revenues in the 22 aftermath of natural disasters. Environment and Development Economics, 16 (1), pp.113- 128. Raddatz, C., 2007. Are External Shocks Responsible for the Instability of Output in Low-Income Countries? Journal of Development Economics, vol. 84, pp. 155-187. Panwar, V. and Sen, S., 2019. Economic impact of natural disasters: An empirical re-examination. Margin: The Journal of Applied Economic Research, 13 (1), pp.109-139. Rasmussen, T., 2004. Macroeconomic implications of natural disasters in the Caribbean. Restuccia, D., Yang, D.T. and Zhu, X., 2008. Agriculture and aggregate productivity: A quantitative cross-country analysis. Journal of Monetary Economics, 55 (2), pp.234-250. Romer, C.D. and Romer, D.H., 2015. New evidence on the impact of financial crises in advanced countries (No. w21021). National Bureau of Economic Research. Skidmore, M. and Toya, H., 2002. Do natural disasters promote long-run growth?. Eco- nomic inquiry, 40 (4), pp.664-687. Stern, N.H., 2007. The economics of climate change. ABC. Strulik, H. and Trimborn, T., 2019. Natural disasters and macroeconomic perfor- mance. Environmental and Resource Economics, 72(4), pp.1069-1098. Teulings, C.N. and Zubanov, N., 2014. Is economic recovery a myth? Robust estima- tion of impulse responses. Journal of Applied Econometrics, 29 (3), pp.497-514. 23 Appendix The agricultural sector is particularly sensitive to extreme weather events and changes to temperature and precipitation patterns, which is reflected in the positive correlation between countries’ agricultural sector size and their climate vulnerability. Figure 6: Climate risk and agricultural sector Notes: The sample includes 128 countries in 2020, with higher values representing higher risk. Source: IMF Climate Change Dashboard and World Bank WDI. In addition, specification (1) is extended by the term Di,t × Ai,t−1 that captures the in- teraction between disaster intensity and the size of the agricultural sector in the economy. Obtained results suggest that economies with a larger dependence on agriculture indeed experience more severe effects.. Agrarian economies usually have a less developed physical infrastructure, less developed capital markets, and less access to international financial markets. Although numerous variables are included to control for these characteristics, there might still be omitted structural factors that explain why the economy per se is more susceptible to any unexpected shock. Table 5: Interaction regression results Notes: Table shows the estimated interaction coefficients from the panel regressions with the agricultural sector size considered. Source: Author’s calculations. 24