Do Sovereign Catastrophe Bonds Improve Fiscal Resilience? An Application of Synthetic Control Method to Mexico

Natural disasters exert a significant impact on government finances. Catastrophe bonds (CAT bonds) constitute risk-transfer instruments that could help governments improve their fiscal resilience when catastrophic events occur. However, given the very limited issuance of sovereign CAT bonds so far, their actual impact on fiscal resilience is difficult to quantify. There is no literature on this topic currently available. I attempt to fill this gap and assess the impact of CAT bond payouts on the fiscal balance of the Mexican government using the synthetic control method. As an early adopter and repeated issuer of sovereign CAT bonds since 2006, Mexico received its first CAT bond payout in 2017. The payout was triggered by a high-magnitude earthquake that stroke the country in September 2017, with an estimated impact of 0.24% of Mexico’s gross domestic product (GDP). I identify 12 countries that experienced natural disasters with a similar impact on GDP in 2017, but which unlike Mexico have not received a CAT bond payout that year. I then compare post-2017 fiscal balances for Mexico with a synthetic control unit that combines the characteristics of the 12 similar but untreated countries, while controlling for other factors that could have had an impact on this fiscal variable. I find a positive and statistically significant impact of the 2017 CAT bond payout on Mexico’s fiscal balance compared to its synthetic control unit. A series of placebo studies and robustness tests confirm the validity of these findings.


Introduction
Natural disasters have a significant impact on government finances, as they entail direct and indirect costs that can be substantial.As regards direct costs, these typically include costs related to emergency response, relief and recovery efforts, as well as costs associated with the reconstruction of damaged infrastructure.While more difficult to quantify, indirect costs stem from lost revenue from disrupted economic activity, increased demand for social services, and long-term economic impacts emanating from declining potential output, to name just a few.Natural disasters can also negatively impact a government's ability to issue debt instruments, as investors may require higher risk premia.
Catastrophe bonds (CAT bonds) are a type of insurance-linked security that allows issuers to partially transfer natural disaster risk to capital markets.CAT bonds can be a particularly useful financing tool for governments, who are legally and/or morally obliged to provide financial assistance to the affected population and repair damaged public infrastructure in the aftermath of a natural disaster.However, only a few countries have so far issued CAT bonds.Mexico is an early adopter, being present in the CAT bond market since 2006.Through their design, CAT bonds trigger payouts to the issuer if a pre-determined natural disaster event occurs during the tenure of the bond.In 2017, Mexico became the first country to receive a CAT bond payout, after a high-magnitude earthquake.The earthquake occurred in September 2017 and caused damage estimated at 0.24% of Mexico's 2017 gross domestic product (GDP).Peru and the Philippines are the only other two countries to have received CAT bond payouts, in 2019 and 2022, respectively.However, the recency of the latter two payouts does not allow for an accurate quantification of their impact using causal inference.I therefore focus on the Mexican case.
Due to the scarcity of sovereign CAT bond issuance, quantifying the impact of these financial instruments on fiscal outcomes is challenging.There is currently no literature dealing with this topic.This paper aims to fill the gap in the literature by assessing the impact of CAT bond payouts on fiscal balances using the synthetic control method (SCM), which is an appropriate research method for settings with one treated unit and multiple control units.For the reasons outlined above, Mexico is the only country that meets the criteria for being considered the treated unit in the SCM settings.I identify the 12 countries that in 2017 experienced a natural disaster with a similar impact on their respective GDP as the earthquake in Mexico, but which unlike Mexico, have not received a CAT bond payout.The 12 countries in the "donor pool" are: Australia, Bangladesh, Chile, Costa Rica, Croatia, Iran, Madagascar, the Philippines, South Africa, Sri Lanka, Thailand, and the United States of America (United States).I then compare the difference in government fiscal balance as a percentage of GDP before and after the 2017 CAT bond payout for the treated unit -Mexico -to a weighted combination of the 12 similar but untreated (i.e., no CAT bond payout in 2017) countries.Overall, I document a positive and statistically significant impact of the CAT bond payout on Mexico's fiscal balance ratio.More precisely, I find that Mexico's fiscal balance-to-GDP ratio in 2017 was 4 percentage points higher compared to that of its synthetic counterpart.In other words, the Mexican government achieved a fiscal balance ratio that was 4 percentage points higher compared to a counterfactual situation where no CAT bond payout had been received in 2017.The validity of these results is confirmed across a series of placebo studies and robustness tests.
The paper makes a significant contribution to the literature on CAT bonds and, more broadly, to the disaster risk finance literature.First, it provides empirical evidence on the impact of CAT bond payouts on the fiscal balance of Mexico -an emerging economy that is highly exposed to natural hazard risk.To the best of my knowledge, this is the first paper to provide quantitative evidence on the benefits of CAT bonds from a sovereign issuer's perspective.Second, a novelty of this study is the use of the SCM method to assess whether CAT bond payouts have had an impact on Mexico's fiscal balance, while controlling for other relevant factors.This method has been deployed to study the economic impact of the German reunification and of California's tobacco control policy, inter alia, but there is however no guidance on how to apply the SCM to evaluate the fiscal impact of sovereign CAT bonds.My paper therefore represents an important first step in filling this gap.
The rest of the paper is organized according to the following structure.Sect."Literature review" briefly reviews the literature on the fiscal impacts of natural disasters and discusses the challenges of quantifying the impact of CAT bonds on fiscal outcomes.Sect."Data and methodology" provides the background to the analysis and describes the data and the methodology.Sect."Results and discussions" discusses the results obtained from the SCM and rolls out placebo studies to assess the statistical significance of the results.It also presents a series of robustness tests.Sect."Conclusions" concludes the paper.

The Fiscal Impacts of Natural Disaster Events
The occurrence of a natural disaster event can lead to changes in both government expenditures and revenues.First, natural disasters may result in future expenditures that governments are obliged to make, very often because they are required to do so by law and sometimes because they are morally obliged to intervene.These constitute a government's disaster-related contingent liabilities that may become actual expenditures when disaster strikes.Second, there are two main channels through which natural disasters can have an impact on government revenues.For instance, disasters can negatively impact personal and corporate income, public and private consumption and the extraction of natural resources, leading to a reduction in the tax base.Additionally, a government may resort to deliberate tax cuts to alleviate the pressure on households and businesses, thus further reducing tax revenue.The loss of tax revenue creates considerable fiscal risks to governments (OECD/ The World Bank 2019; Marulanda et al. 2022).
In the category of disaster-related contingent liabilities, a distinction can be made between explicit and implicit contingent liabilities.The former can be defined as "payment obligations based on government contracts, laws or clear policy commitments that could fall due in the event of disaster" (OECD/The World Bank 2019).This category encompasses, for example, expenses incurred from the destruction of public buildings and infrastructures, as well as those resulting from pre-arranged commitments.Reconstruction costs tend to be larger in developing countries (Adam and Bevan 2020).
Conversely, disaster-related expenses that the government incurs without a prior formal obligation are referred to as implicit contingent liabilities.These expenses may be the result of a government's effort to safeguard the economy from recession or be prompted by political or ethical considerations.Implicit liabilities are more difficult to identify and measure compared to explicit contingent liabilities.While the government can use past disaster-related expenses as a benchmark for estimating potential government spending in the aftermath of a disaster, it may be challenging to accurately forecast those contingent liabilities stemming from pure political and ethical considerations (OECD/The World Bank 2019).An example of liabilities arising from political and/or ethical considerations could be the need for government intervention to prevent income inequalities from further increasing.For instance, as shown by Pleninger (2022) in a study of United States (US) households, natural disasters have an asymmetric effect on incomes, the negative impact being stronger on middle incomes.Governments therefore typically provide post-disaster assistance to the lower-income fringes of the population.
In addition to government expenditures and risks to revenues arising from the occurrence of natural disasters, a number of potential indirect disaster-related fiscal impacts can be identified.These impacts are more difficult to observe, but they can be just as significant as the direct ones.These include the loss of economic activity, as businesses in the disaster-affected areas may be obliged to halt their activities temporarily or even permanently, leading to job losses and lower economic output.This can have ripple effects on the economy, as income losses will lead to subdued consumer spending and further job losses.Empirical evidence shows that private consumption tends to decline by between 0.98% and 1.54% (i.e., depending on model specification) in the aftermath of a natural disaster (Arouri et al. 2015;Cantelmo et al. 2023).Moreover, as suggested by Cantelmo et al. (2023), the effects of natural disasters may go beyond the short-term impact on consumption, with average annual GDP growth nearly 1% lower in disaster-prone countries compared to nondisaster-prone countries.
Natural disasters can also have an impact on the financial sector.Financial markets may experience volatility in the aftermath of a natural disaster, which could in turn result in higher borrowing costs for the government as investors require higher risk premia.In a recent study of the European sovereign credit default swap (CDS) market, Di Tommaso et al. (2023) show that the CDS market reacts to natural disasters, whereby sovereign risk increases by 0.38% on a 10-day event window following the occurrence of the disaster event.In a similar vein, Mallucci (2022) concludes that hurricane risk limits the ability of governments in Caribbean countries to issue debt.Finally, considering a different market segment, Lanfear et al. (2019) point to strong anomalies in equity stock returns in the US over the period 1990-2017 in response to landfall hurricanes.The negative impact tends to be stronger for growth stocks (i.e., stocks whose valuation depends to a large extent on future prospects).

Quantifying the Benefits of CAT Bonds for a Sovereign Issuer
A CAT bond is an insurance-linked security that allows the cedent to obtain financing from capital markets, if a natural disaster event with pre-defined parameters -for instance, an earthquake with a magnitude above a pre-determined threshold, or wind speed in excess of a pre-determined value -takes place during the tenure of the CAT bond.A growing strand of empirical literature documents the benefits of CAT bonds for sovereign issuers in terms of improving the management of the national budget when a disaster event occurs (Ghesquiere and Mahul 2010; Clarke et al. 2017).Obtaining insurance against natural disasters through CAT bonds can therefore be considered an "adaptation policy for countries with exposure to climate change risks" (Ando et al. 2022).
Quantifying the impact of CAT bonds on sovereign issuers' fiscal resilience can be a very challenging task for several reasons.The first challenge stems from limited data availability, as so far only a few governments have issued CAT bonds.These tend to be countries with high exposure to natural hazard risk, and relatively low levels of public debt and middle-to high income (Maran 2023).Only the governments of Chile (2018, 2023), Colombia (2018), Jamaica (2021), Mexico (2006, 2009, 2017, 2018and 2020), Peru (2018), and the Philippines (2019) have so far issued CAT bonds.
Another challenge stems from the complexity of the CAT bond structure, which often involves multiple risk layers, each layer with different triggers and payout structures.In an influential paper, Lee and Yu (2002) highlight the numerous factors that influence CAT bond prices, namely: moral hazard, basis risk, catastrophe occurrence intensity, loss volatility, trigger level, the capital position of the issuer, the debt structure, as well as interest rate uncertainty.As shown by Zhao and Yu (2020) in a more recent empirical study, complex CAT bonds -defined for the purpose of that study as "bonds that provide coverage for multiple types of perils and multiple territories" -tend to have longer maturities, higher spreads, as well as greater expected loss, compared to less complex CAT bonds.In a study of CAT swap contracts, Lo et al. (2021) similarly argue that the pricing of this type of contract is highly complex as it will not only depend on the amount of catastrophic losses -subject to a great level of uncertainty, but also on the type of CAT bond trigger.CAT bond liquidity is another factor that influences prices.Zhao and Yu (2019) conclude that, between 2002 and 2016, the CAT bond liquidity premium accounted for approximately 10% of the average CAT bond spread in the secondary market during that period.The authors additionally show that severe events such as natural hazards or the 2008 global financial crisis tend to exert significant upward pressure on CAT bond liquidity premia.
Finally, identifying a suitable quantitative method for modelling CAT bond contracts could be equally challenging.Chang et al. (2011), for instance, compare several pricing methods and conclude that, for the specific case of CAT equity put options, Markov-Modulated Poisson Processes tend to perform better than competing methods such as the exponential growth pattern.For their part, Lee and Yu (2002) use the Monte Carlo method to compute the prices of default-free and default-risky CAT bonds.
In the theoretical literature, Chang et al. (2020) develop a sequential optimization framework with two agents and consider the impact of various CAT bond characteristics on insurers/reinsurers' portfolio allocation strategies.This study corroborates the results from the empirical literature by pointing out that the optimum portfolio allocation is a complex undertaking that needs to take several parameters into account (e.g., collateral type, multiple trigger types, number of tranches, payment flows, etc.) It can therefore be concluded based on this relatively rich strand of empirical and theoretical literature that the inherent complexity of CAT bond contracts poses significant challenges to the assessment of a CAT bond's impact on an issuer's financial position.
Finally, and perhaps most importantly, the impact of CAT bonds on public finances may be difficult to distinguish from the influence of other factors that may affect a government's fiscal position.The literature dealing with the determinants of fiscal balances is vast.The explanatory variables most commonly cited in the economic literature can be grouped into four categories, namely -fiscal, macroeconomic, political and other variables.The impact of CAT bonds needs to be estimated in conjunction with these numerous factors, which will be discussed in detail below.
As regards the fiscal determinants of the fiscal balance, interest payments and fiscal rules targeting budget balances are commonly cited.Interest payments on public debt exert a significant impact on the fiscal balance.Higher interest payments are associated with an increase in the level of public debt, which can then create a strain on the fiscal balance (Lora and Olivera 2007).As regards fiscal rules, a broad strand of literature confirms the significant positive influence that the existence of these rules exerts on fiscal balances (Maltritz and Wüste 2015;Caselli and Reynaud 2020;Gootjes and De Haan 2022).In particular, fiscal rules that target the budget balance and apply to the general government are shown to increase the soundness of public finances (Chrysanthakopoulos and Tagkalakis 2023).
In the category of macroeconomic variables, the unemployment rate and output growth impact the fiscal balance mainly through automatic stabilizers, via tax revenues and unemployment-related expenditures.When unemployment rises, government spending on unemployment benefits automatically increases, which helps stimulate the economy, reduce the negative impact on revenues and in turn on the fiscal balance.Conversely, when GDP growth is strong and the rate of employment is low, government spending on unemployment benefits declines, limiting inflation and its impact on the fiscal balance (Dolls et al. 2012;Huang and Yang 2021).Real GDP per capita is another potential explanatory variable, being positively correlated with the fiscal balance.In other words, higher (lower) levels of per capita GDP will result in higher (lower) levels of revenue that the government can collect from taxes, and thus turn into either a budget surplus or deficit (Mawejje and Odhiambo 2022).Inflation is also often included among the macroeconomic variables affecting the evolution of the fiscal balance.It may have an automatic effect on government expenditures through higher spending on social security programmes aimed at helping citizens cope with the rising cost of living.Another automatic effect stems from the increase in tax revenues, as income rises with inflation.Inflation can also trigger a decrease in the value of government bonds and other debt instruments, which in turn may lead to an increase in the cost of borrowing for the government (Chu et al. 2011).In addition, including inflation as an explanatory variable may be warranted if governments are assumed to adjust policies in case of rising inflation, to safeguard competitiveness and alleviate pressures on fixed exchange rates for countries that participate in an exchange rate agreement (Yang et al. 2018).
Regulatory quality and the control of corruption represent some of the most widely cited political variables.Indeed, many studies on the determinants of fiscal balances focus on regulatory quality as an explanatory variable.The idea is that high regulatory quality is associated with effective and efficient regulations, a better allocation of resources and improved revenue collection.In turn, this helps to improve the government's fiscal balance.For example, independent fiscal institutions are shown to exert a positive and significant influence on government fiscal balances in the European Union (EU) member countries, resulting in smaller budget deficits (Giesenow et al. 2020;Căpraru et al. 2022).It has also been suggested in the literature that corruption -associated with phenomena such as tax evasion, a large size of the shadow economy, irregularities in procurement, theft or bribes -is detrimental to sound public finances, as it decreases state revenues and/or promotes a wasteful spending of revenues (Blackburn and Powell 2011), leading to an increase in budget deficits (Oto-Peralias et al. 2013).
The last category encompasses all other variables that could have an impact on fiscal balances and cannot be grouped under any of the previous three categories.Among the plethora of other variables, population ageing, and systemic banking crises are some of the most popular explanatory variables.Population ageing exerts a large direct impact on budgetary expenditures and revenues.As the population ages, the number of people who are retired and receive government benefits increases, while the number of people working and paying taxes declines.This puts a strain on government finances, as supporting the number of retirees requires higher spending.Additionally, healthcare costs also tend to increase with an ageing population, adding further pressure on the fiscal balance (Kitao 2015;Cho and Lee 2022).The banking crisis literature broadly concludes that the occurrence of a crisis with systemic implications is associated with a sharp deterioration of the fiscal balance.In case a systemic banking crisis occurs, government intervention may be necessary to prevent widespread bank failures.This can lead to a significant increase in government spending.In addition, a banking crisis that spreads to the real economy may lead to a decrease in tax revenue, as economic activity slows down and households and businesses struggle to stay afloat (Honohan and Klingebiel 2003;ECB 2015).
In this paper, I try to overcome these challenges and fill the gap in the literature by quantifying the impact of CAT bonds on fiscal outcomes using the SCM method and Mexico as a case study.To the best of my knowledge, at the time of writing my paper is the first one to deal with the important topic of measuring the benefits of CAT bonds from a sovereign issuer's perspective and thus makes an important contribution to the literature.

Background
The typical CAT bond structure involves three main actors, namely: (i) the entity that is exposed to natural disaster risk and seeks to transfer part of this risk to the capital market -known as the "sponsor"; (ii) the Special Purpose Vehicle (SPV); and (iii) the investors.The sponsor enters into an insurance agreement with the SPV.In exchange for the premiums paid by the sponsor, the SPV issues the CAT bonds to investors and subsequently invests the proceeds from the issuance in highly rated securities, which are held in a collateral trust.The SPV is also responsible for making regular coupon payments to the CAT bond investors.The typical CAT bond maturity is three years.If a pre-determined natural disaster event takes place before the stated maturity date, the SPV issues an insurance payout to the sponsor.Conversely, if the pre-determined event does not occur during the bond term, the assets held in the collateral trust are liquidated and the sale proceeds are distributed to the CAT bond investors.
The World Bank plays a significant role in the issuance of CAT bonds by governments in emerging market economies (EMEs).It acts as a facilitator and advisor in the issuance process.It also provides technical expertise and guidance to governments in EMEs to help them assess their exposure to natural disasters and develop risk management strategies.This includes conducting risk assessments, developing disaster risk financing frameworks and structuring the CAT bond transactions.Furthermore, the World Bank plays a crucial role in attracting investors to participate in these CAT bond transactions.It thus helps governments in EMEs to access the global capital markets and connect with institutional investors that have an investment interest in CAT bonds (World Bank 2020).The CAT bonds issued with the help of the World Bank thus do not necessitate an SPV, as the World Bank assumes this role.
Mexico is highly exposed to natural disasters and is one of the EMEs that benefitted the most from World Bank support in bringing its sovereign CAT bonds to the market.Mexico first issued CAT bonds in 2006, becoming the first sovereign to use this type of financial instrument.It then returned to the market with additional issuances, the most recent one being in 2020.All of these transactions were arranged by the World Bank within the framework of its "MultiCat" Programme.
The structure of the July 2017 transaction, which will be the focus of this paper, is shown in Fig. 1.The 2017 CAT bond provides the Government of Mexico with USD 360 million (United States dollars) insurance coverage for earthquakes and named Atlantic and Pacific storm events, of which USD 150 million were earmarked for the earthquake tranche.The scheduled maturity date for the earthquake tranche is 11 August 2020.The Government of Mexico pays premiums to the World Bank, at a rate of 4.50% per annum.Finally, the yearly coupon rate for this tranche is determined according to the formula: 6-month USD London interbank offered rate (LIBOR) + 4.12%, subject to a minimum rate of 4.50% (World Bank 2017).
Less than three months after the issuance, Mexico received a payout from the 2017 World Bank-supported CAT bond transaction.This was the first payout received since the country began issuing CAT bonds in 2006.The payout was triggered by an earthquake with magnitude 8.1 on the Richter scale.The earthquake took place in the Mexican states of Oaxaca, Chiapas and Tabasco on 7 September 2017.According to figures from the EM-DAT database (EM-DAT 2023), the total damage caused by this earthquake amounted to approximately 0.24% of Mexico's GDP in 2017.In the aftermath of the earthquake, the Mexican government confirmed that it would receive a payout of USD 150 million from the World Bank-supported 2017 CAT bond after the calculation agent confirmed that the earthquake on 7 September 2017 met all the parameters required to trigger and default the earthquake tranche of the CAT bond notes (Artemis 2017).
In light of the above, the purpose of this study is to assess whether the 2017 CAT bond payout has had any fiscal impact for the Mexican government.In general, if a natural disaster event occurs and triggers a CAT bond payout, the government's fiscal balance could be affected in several ways.I posit as a null hypothesis that the CAT bond payout should have a positive short-run impact on the fiscal balance, as it can help alleviate the immediate financial burden on the government and enable it to quickly mobilize resources to assist affected areas and populations without the need to borrow additional funds from the banking sector or capital markets.CAT bonds indeed offer much faster payouts compared to traditional insurance contracts, with funds typically disbursed to the issuing government within a few weeks after the occurrence of the natural disaster event (World Bank 2021).
On the other hand, the sovereign CAT bond payout may not be substantial enough to cover the costs of disaster response, recovery, and reconstruction.In addition, as often argued in the literature, the high costs associated with issuing CAT bonds (Michel-Kerjan et al. 2011;Ando et al. 2022) could offset their benefits.In addition to the transaction costs, which are nevertheless lower due to the support provided by the World Bank compared to the typical CAT bond, the issuing government also needs to pay premiums to the World Bank.For the 2017 transaction, the premium rate was set at 4.50%, which is higher than the 3.3% average interest rate paid by the Government of Mexico on its public debt between 2002 and 2016.The CAT bondrelated costs can put pressure on a government's fiscal balance, especially if the respective government already has a high debt stock and/or limited fiscal capacity.I therefore state as an alternative hypothesis that the 2017 CAT bond payout has no or negative impact on the fiscal balance of the Mexican government because the associated costs offset the benefits.

Data Description
This paper uses the SCM methodology, which requires data on one treated unit and multiple control units.In selecting a treated unit, a first step is to look at countries that have received a CAT bond payout in the past.In addition to the 2017 CAT bond payout to Mexico discussed in detail in Sub-section "Background", I document that only two other countries have received such payouts: Furthermore, the SCM method requires data for a sufficiently long period before and after the event whose effect is to be investigated.Since the Peru and the Philippine CAT bond payouts are relatively recent, I focus on the 2017 Mexican CAT bond as a basis for the case study, thus considering Mexico as the treated unit in the SCM.Owing to data availability, the sample period stretches from 2002 to 2021 and the frequency is yearly.
The SCM requires control units that have similar characteristics with the treated unit.I therefore restrict the sample of units in the donor pool to countries that, like Mexico, experienced a natural disaster in 2017, and whose estimated impact ranges from 0.1% to 0.6% of GDP (the impact of the earthquake that triggered the CAT bond payout to the Mexican government in 2017 was estimated at 0.24% of 2017 Mexico GDP).To obtain the list of control units based on this criterion, I take the list of natural disasters reported in the EM-DAT database (EM-DAT 2023).The database references multiple types of natural disasters, of which I retain the climatological, geophysical, hydrological and meteorological ones, since CAT bond coverage is typically limited to these types of hazards.The EM-DAT database considers several criteria in order to decide whether a disaster should be included in the database or not (at least one of the criteria must be fulfilled): (1) 10 or more people deaths have been recorded; (2) 100 or more people have been affected, injured, or left homeless; or (3) the country has declared a state of emergency and/or has launched an appeal for international assistance (EM-DAT 2023).
Although Peru would have qualified for inclusion in the donor pool, this country received a CAT bond payout in 2019 and cannot therefore be considered.On the other hand, even though it has received a CAT bond payout in 2022, the Philippines can be included in the sample since the payout falls outside of the 2002-2021 span.As such, I identify 12 countries that can be included in the donor pool: Australia, Bangladesh, Chile, Costa Rica, Croatia, Iran, Madagascar, the Philippines, South Africa, Sri Lanka, Thailand, and the United States.Table 1 provides the list of countries in the donor pool with the corresponding natural disaster that took place in 2017.
The selection of the dependent fiscal variable is another important step in the empirical analysis.I will focus on nominal budget balance data as dependent variable rather than on government debt data.I make this choice because public debt represents a stock, while the budget balance constitutes a flow.Governments typically define their budgetary targets in flow terms rather than in stock terms, as stock variables are more affected by factors outside of government control (i.e., economic growth, exchange rate developments or As discussed in Sub-section "Quantifying the benefits of CAT bonds for a sovereign issuer", the impact of CAT bond payouts on Mexico's fiscal balance needs to be assessed in conjunction with several other factors that influence this variable.Based on a review of the literature, the factors explaining changes in the fiscal balance can be grouped under four broad categories, namely fiscal, macroeconomic, political, and other variables.Table 2 summarizes the independent variables that I use to explain the evolution of the fiscal balance within the SCM framework.
The next step is to apply the SCM approach, by following the steps outlined in subsection "Synthetic control method: theoretical background".

Synthetic Control Method: Theoretical Background
Comparative case studies have been extensively applied to evaluate the causal effect of large-scale events.In some more recent papers, Chen et al. (2007) use Fujian and Taiwan (China) in a comparative case study of the driving forces for cultivated land changes over time, while Neumann et al. (2022) focus on eight public organizations based in Switzerland to explore the drivers behind artificial intelligence adoption in public organizations.These examples notwithstanding, comparative case studies of this type have several shortcomings.One of these limitations is that the selection of the control units is not performed in a formal manner, but instead it relies on informal assessments of similarity between the treated units and these control units.Additionally, when there are only a small number of units of observation available, for instance in studies involving countries or regions, no single unit alone may provide a good comparison for the treated unit.
The SCM posits that, when the units of observation are a small number of aggregate entities like for instance countries or regions, a combination of unaffected units often provides a more appropriate comparison than any single unaffected unit alone.The SCM aims to formalize the selection of the comparison units using a data-driven procedure (Abadie and Gardeazabal 2003;Abadie et al. 2010;Abadie 2021).For this reason, the SCM method has become increasingly popular in empirical research applied to the field of economics.
I obtain data for J + 1 units (i.e., countries): j = 1, 2, … , J + 1 .Without loss of general- ity, the SCM assumes that the first unit (j = 1) is the treated unit (i.e., Mexico).The first unit, namely Mexico in my paper, is the unit affected by the intervention of interest -the payout triggered by the CAT bond in 2017.The "donor pool", that is, the set of potential control units, j = 2, … , J + 1 comprises a set of untreated units, that have not received a CAT bond payout in 2017.The SCM also assumes that data span T periods and that the first T 0 periods are before the intervention.For each unit j and time t, the outcome of interest, Y jt, -the fiscal balance-to-GDP ratio in my paper -can be observed.For each unit j, one also observes a set of k predictors of the outcome, X 1j , … , X kj , which are themselves unaffected by the intervention (refer to sub-section "Data description" above for the list of predictors).The k × 1 vectors X 1 , … , X J+1 contain the values of the predictors for units j = 1, 2, … , J + 1 , respectively.The k × j matrix, X 0 = X 2 … X J+1 , collects the values of the predictors for the J untreated units -in this paper, these are the predictors for the 12 Dummy variable that takes a value of 1 if a systemic banking crisis occurred and 0 otherwise Laeven and Valencia (2018) countries included in the donor pool.For each unit j and time period t, Y N jt is defined as the potential response without intervention (i.e., in the absence of the CAT bond payout).For the unit affected by the intervention, j = 1, and a post-intervention period t > T 0 , Y I 1t represents the potential response under the intervention.For easier reference, where applicable, I will replace the subscript 1 with MEX in the following explanatory paragraphs.
The effect of the intervention of interest for the affected unit in period t (with t > T 0 ) -namely, the effect of the 2017 CAT bond payout on Mexico's fiscal balance, is: Because the treated unit is exposed to the intervention after period T 0 , it follows that for t > T 0 , Y MEXt = Y I MEXt .In other words, for the unit affected by the intervention and a postintervention period one can observe the potential outcome under the intervention.Estimating Y N MEXt for t > T 0 poses major challenges, as it entails determining how the outcome of interest would have evolved for the treated unit Mexico in the absence of the intervention.As is evident from Eq. ( 1), given that Y I MEXt is observed, the problem of estimating the effect of the intervention is equivalent to the problem of estimating Y N MEXt .Moreover, Eq. ( 1) allows the effect of the intervention to change over time, as the effects of the intervention may not be contemporaneous and may accumulate or dissipate over time.
As regards the estimation of Y N MEXt , comparative case studies would aim to reproduce the value of Y N MEXt -that is, the value of the outcome variable that would have been observed for the affected unit in the absence of the intervention by using one unaffected unit or a small number of unaffected countries that had similar characteristics to Mexico at the time of the intervention.However, when the data consist of a few aggregate entities, such as countries, finding a single unaffected unit that provides a suitable comparison for the unit affected by the intervention of interest is often challenging.As previously mentioned, the SCM posits that a combination of units in the donor pool may approximate the characteristics of the affected unit considerably better than any unaffected unit alone.A synthetic control is defined as a weighted average of the units in the donor pool -in this paper, the synthetic control is the weighted average of the 12 countries in the donor pool that have similar characteristics to Mexico.A formal representation of a synthetic control could be a J × 1 vector of weights, W = w 2 , … , w J+1 � .The synthetic control estimators of Y N MEXt and MEXt are, respectively, given by: and In order to avoid extrapolation, the weights are restricted to be non-negative and to sum to one.These two restrictions yield synthetic controls that are weighted averages of the outcomes of units in the donor pool, with typically sparse weights.This means that only a small number of units in the donor pool contribute to the estimation of the counterfactual of interest, ŶN MEXt , and the contribution of each unit is represented by its synthetic control weight.Since synthetic control weights are defined as weighted averages, a synthetic control counterfactual estimate is particularly more transparent relative to alternative causal inference methods (Abadie 2021).
Expressing the comparison unit as a synthetic control raises questions around the practical choice of the weights, w 2 , … , w J+1 .Abadie and Gardeazabal (2003), Abadie et al. (2010) and Abadie (2021) propose selecting w 2 , … , w J+1 in such a manner that the result- ing synthetic control best resembles the pre-intervention values of predictors of the outcome variable for the treated unit.That is, given a set of non-negative constants, v 1 , … , v k , Abadie and Gardeazabal (2003), Abadie et al. (2010) and Abadie (2021) propose to select the synthetic control W * = w * 2 , … , w *

J+1
, that minimizes the following: Subject to the restriction that the weights w 2 , … , w J+1 are greater than 0 and sum to one.It then follows that the estimated treatment effect for the treated unit at time t = T 0 + 1, … , T is given by: The positive constants v 1 , … , v k in Eq. ( 4) reflect the relative importance of the syn- thetic control replicating the values of each of the k predictors for the treated unit, P 1MEX , … , P kMEX .For a given set of weights v 1 , … , v k , minimizing the difference expressed in Eq. ( 4) can be easily accomplished using constrained quadratic optimization.More specifically, each potential choice of which can be determined by minimizing the difference in Eq. ( 4), subject to the restriction that the weights in W(V) are positive and sum to one.In order to choose V, a simple selector of v h is the inverse of the variance of P h1 , … , P hJ+1 , which in fact rescales all rows of P 1 ∶ P 0 such that they have unit variance.

Results from the Baseline Specification
In this section I report the results obtained from the baseline SCM approach.The treated unit is Mexico, while the synthetic control unit combines the characteristics of the 12 countries in the donor pool by applying the procedure outlined in Sub-section "Synthetic control method: theoretical background".The treatment intervention is the 2017 CAT bond payout received by Mexico in the aftermath of a high-magnitude earthquake.
The first step in the discussion of results is to look at the weights assigned to the 12 countries that constitute the donor pool.These weights play a critical role in determining the validity and accuracy of the estimated treatment effect.From an economic perspective, the assignment of weights in the SCM can be interpreted as a reflection of the economic relevance or similarity of the different control units to the treated unit.The goal is to construct a synthetic control unit that closely resembles the treated unit -Mexico -in terms of pre-treatment characteristics and trends.Therefore, countries that are more economically similar to Mexico are assigned higher weights, indicating that they have a greater influence on the construction of the synthetic control unit.Conversely, countries that largely differ (4) from Mexico in terms of economic structures and trends are assigned lower or even zero weights.
Table 3 reports the weights of each of the 12 countries in the donor pool, which combined constitute a synthetic control unit for Mexico.Costa Rica receives the highest weight (0.377), followed by the Philippines (0.341) and Madagascar (0.216).It means that Costa Rica, the Philippines, and Madagascar are the three countries in the donor pool that are most similar to Mexico in terms of economic structures and dynamics, therefore being more likely to respond similarly to the treatment.Costa Rica, in particular, is very similar to Mexico in terms of average interest paid in the pre-2017 period (3.56% versus 3.33% in Mexico), average inflation rate (8.34% versus 8.27%), but also in terms of the banking crisis (0.03 both) and budget balance fiscal rule (0.72 versus 0.55) variables.
Albeit to a much lesser extent, Croatia (0.045), Sri Lanka (0.019) and Thailand (0.002) also contribute to the synthetic control unit.All the remaining countries in the donor pool have zero weights.It can therefore be concluded that the synthetic version of Mexico is a weighted average of Costa Rica, the Philippines, Madagascar, Croatia, Sri Lanka, and Thailand.It is not surprising that Mexico has more affinities with the middle-income economies included in the donor pool rather than with high-income countries such as Australia, Chile and the United States.
Table 4 compares the pre-2017 characteristics of Mexico to those of the synthetic control unit that closely matches the characteristics of Mexico.For comparison purposes, I also report the sample average for all the 12 countries in the donor pool, including those countries that received zero weights.It can be noticed that the synthetic control unit is much more similar to Mexico than the average of all the countries in the donor pool.Mexico and its synthetic control are very similar in terms of pre-2017 interest payments, unemployment, proportion of the elderly population and occurrence of a systemic banking crisis.Although the discrepancies are slightly larger, Mexico is also more similar to its synthetic control than to the entire sample as regards pre-2017 budget balance fiscal rule, inflation and control of corruption.Conversely, Mexico is more like the entire sample in terms of pre-intervention output growth, per capita GDP and regulatory quality.The large difference in terms of per capita GDP can be explained by the fact that Mexico displays a much higher mean value for this indicator in the pre-2017 period compared to the middle-income countries that compose its synthetic control unit.
Figure 2a compares the trajectory of Mexico's fiscal balance-to-GDP ratio before the CAT bond payout and an unweighted average of the countries in the donor pool, for the years 2002-2016.It can be noticed that an unweighted average of the fiscal balance ratio for the countries in the donor pool fails to reproduce the trajectory of the fiscal balance ratio for Mexico even before the CAT bond payout takes place in 2017.Conversely, Fig. 2b reports the trajectory of the fiscal balance-to-GDP ratio for Mexico and for a synthetic control calculated in the manner explained in Sub-section "Synthetic control method: theoretical background".This figure shows that a weighted average of the 12 countries in the donor pool more closely matches the trajectory of Mexico's fiscal balance ratio during the 2002-2016 period, before the CAT bond payout was triggered.I estimate the effect of the CAT bond payout to Mexico as the difference between the fiscal balance-to-GDP ratio for Mexico versus its synthetic control unit after the CAT bond payout was triggered in 2017.Figure 3 shows that as soon as Mexico received the payout in 2017, the fiscal balance ratios have started to diverge noticeably.While the fiscal balance ratio continued to deteriorate for the synthetic version of Mexico, remaining in the deficit area, the actual Mexico experienced a considerable improvement in its fiscal balance ratio, which turned positive in 2017 and remained in surplus territory for most of the 2017-2021 period.Figure 3 therefore suggests that the 2017 CAT bond payout received by the Mexican government in the aftermath of a high-magnitude  earthquake had a positive and substantial effect on Mexico's fiscal balance compared to countries that experienced a similar natural disaster event in 2017 but have not received a CAT bond payout.Another way to assess the impact of the 2017 CAT bond payout to Mexico is to look at the gap between fiscal balance ratios between Mexico and its synthetic version before and after the payout.The gaps are shown in Fig. 4. Prior to the 2017 payout, the gap between the fiscal balance ratios was for the most part unfavourable to the actual Mexico compared to its synthetic control, as shown by the negative values of the gap.For instance, in 2014 Mexico's fiscal balance-to-GDP ratio was nearly one percentage point lower than the corresponding ratio of its synthetic version.Conversely, when the CAT bond payout was triggered in 2017, the fiscal balance-to-GDP ratio of Mexico inched 4 percentage points higher compared to that of its synthetic version, the largest positive discrepancy observed during the entire sample period.For comparison, the average gap in fiscal balances between Mexico and its synthetic version stood at about 0.3 percentage points during the pre-payout period 2002-2016.Moreover, the Mexican government continued to post fiscal surpluses in the aftermath of the payout, while Mexico's synthetic version experienced fiscal deficits during 2017-2021 (Fig. 3).Overall, these results suggest that the CAT bond payout exerted a positive impact on Mexico's fiscal balance as a share of GDP, thus confirming the null hypothesis posited in Sub-section "Background".

Inference about the Effect of the 2017 CAT Bond payout to Mexico
In a similar vein to Abadie et al. (2010) and Abadie et al. (2015), I run placebo tests to determine whether the effect of the 2017 CAT bond payout to Mexico is statistically significant as opposed to being driven entirely by chance.More specifically, the idea behind placebo tests is to iteratively apply the same SCM methodology to each of the 12 countries in the donor pool, assuming that the CAT bond payout would be disbursed to other countries that in reality have not received such a payout.Placebo tests provide an assessment of whether the estimated treatment effect for Mexico is large relative to the effect estimated for a country that did not receive a CAT bond payout in 2017.Put differently, the placebo approach rests on the idea that, if one observes treatment effects of similar or even greater magnitude in cases where the intervention did not occur, confidence in a specific synthetic control estimate would be significantly weakened (Abadie et al. 2015).
I will therefore consider the effect of the 2017 CAT bond payout to Mexico to be significant if the estimated effect for this country is large compared to the distribution of placebo effects.To do so, I compute the country-specific root mean squared prediction error (RMSPE), 1 which in this paper measures the magnitude of the gap between the fiscal balance-to-GDP ratio between each country and its synthetic version.As in Abadie et al. (2010), Abadie et al. (2015) and Esaka and Fujii (2022), for each country, I compute the RMSPE ratio by dividing its post-intervention RMSPE (i.e., post-2017 RMSPE) to its preintervention RMSPE (i.e., pre-2017 RMSPE).
Figure 5 illustrates the corresponding RMSPE ratios for Mexico and the 12 countries in the donor pool.It can be noticed that Mexico is by far and large the country with the largest RMSPE ratio, which is approximately equal to 22.This ratio has a p-value of 0.0769 and is thus statistically significant at the 10% level.It can be concluded that, for Mexico, the post-2017 gap in its fiscal balance ratio is around 22 times larger compared to the corresponding pre-2017 gap. 1 The RMSPE is a measure of the lack of fit between the trajectory of the outcome variable for the treated country versus its synthetic control unit.For example, the pre-2017 RMSPE for Mexico is given by: Where T 0 is the time when the intervention occurred (i.e., 2017), Y is the outcome variable (i.e., the fiscal balance-to-GDP ratio), and w represent the synthetic control weights.
For a more detailed discussion on the use of RMSPEs in causal inference refer to Abadie et al. (2010) and Abadie et al. (2015).
An alternative to conduct placebo tests is to reassign the treatment to a different period when there was no intervention.This approach is known as "in-time placebo" (Abadie et al. 2015).I therefore rerun the SCM for the case when the CAT bond payout to Mexico is reassigned to 2011.I choose 2011 because that year Mexico experienced an earthquake with a similar magnitude and economic impact to the 2017 one, and no CAT bond payout was received.The results of this in-time placebo test are displayed in Fig. 6.It can be noticed that the trajectories of the fiscal balance-to-GDP ratios of Mexico and its synthetic version do not diverge considerably between 2011 and 2016, suggesting that the fictitious 2011 CAT bond payout exerts a limited effect on Mexico's fiscal balance.This confirms the large and positive gap between the fiscal balances of Mexico and its synthetic counterpart displayed in Fig. 4 is the result of the 2017 CAT bond payout and is not merely due to model lack of predictive power.

Robustness Tests
The results presented in Sub-section "Results from the baseline specification" may be subject to criticism for various reasons.One may argue that the results could be influenced by the sample of countries in the donor pool and thus by their respective weights in the synthetic control unit.For instance, reducing the number of countries that compose the synthetic version of Mexico may negatively impact the synthetic control's goodness-of-fit.I address this concern by repeating the analysis with alternative synthetic control units.More precisely, I alter the composition of the synthetic control unit in the baseline specification by allowing only combinations of 3, 4, 5, 6, 7 and 8 countries.Countries are randomly assigned to the sub-samples.
Figure 7 plots the evolution of the fiscal balance-to-GDP ratio for Mexico and the synthetic control units with different country combinations, namely 3, 4, 5, 6, 7 and 8 randomly selected countries.Except for the seven-country synthetic control unit (Fig. 7e), the remaining synthetic controls depicted in Fig. 7 yield results that are very similar to those illustrated in the baseline specification, namely a gap of approximately 4 percentage points between the fiscal balance ratios of Mexico and its synthetic version (refer to Fig. 4).This analysis confirms the robustness of the results presented in Sub-section "Results from the baseline specification".

Conclusions
Theory suggests that CAT bonds can play an important role in improving the fiscal resilience of governments.By providing a means to partially transfer natural disaster risk to investors, CAT bonds can help governments access capital quickly and reduce their exposure to financial losses caused by natural disasters.This can help governments maintain their fiscal soundness even in crisis times.To test this hypothesis, I apply the SCM to estimate the causal effect of CAT bond payouts in the aftermath of a natural disaster on Mexico's fiscal balance-to-GDP ratio.In this application, Mexico is the treated unit, while 12 other countries similar to Mexico serve as control units.To accurately assess the impact of the CAT bond payout on the fiscal balance of the Mexican government, I identify several other macroeconomic factors that could have influenced this fiscal variable, such as fiscal rules, inflation, unemployment, banking crises, regulatory quality, inter alia.
Overall, the estimation suggests that CAT bonds have exerted a large and positive impact on Mexico's fiscal balance after the 2017 high-magnitude earthquake, which triggered a payout to the Mexican government that same year.In 2017, for instance, Mexico's fiscal balance-to-GDP ratio was 4 percentage points higher compared to that of its synthetic counterpart.For comparison, the average gap in fiscal balances between Mexico and its synthetic version amounted to approximately 0.3 percentage points during 2002-2016.Furthermore, Mexico's fiscal balance ratio remained in positive territory also post-2017, while the 12 countries that constitute its synthetic version experienced fiscal deficits during the same period.The validity of the results is confirmed by a series of placebo studies and robustness tests.These findings indicate that CAT bonds can be an effective tool for governments to manage the financial risks associated with natural disasters.The results have important implications for policymakers and other stakeholders that may have an interest in promoting financial resilience in disaster-prone countries.This paper brings an important contribution not only to the emerging literature on sovereign CAT bonds, but also to the broader literature on disaster risk finance.To the best of the author's knowledge, at the time of writing this is the first attempt to estimate the impact of sovereign CAT bonds on fiscal outcomes.scarcity of the literature is largely due to limited data availability stemming from the still scarce sovereign CAT bond issuance.Moreover, there is currently no guidance on how to apply the SCM approach to the assessment of CAT bonds' impact on fiscal outcomes.This paper can serve as a starting point in filling this gap.As sovereign CAT bond issuance expands, new research avenues will become available and alternative methods for causal inference could be used.For instance, difference-in-difference approaches could also be explored if more countries were to receive CAT bond payouts in the aftermath of a natural disaster event.
In addition, it would be interesting to assess the fiscal implications of sovereign CAT bonds that go beyond the short-term impacts presented in this paper.The impact of the CAT bond payout on the government's fiscal balance will arguably also depend on how the payout is utilized.If the funds are efficiently allocated towards reconstruction and measures to build long-term disaster resilience, there could be a positive long-term impact on the government's fiscal balance.By contrast, if the funds were mismanaged, the government's fiscal challenges may be exacerbated.These constitute additional avenues for future research on the fiscal impacts of sovereign CAT bonds.

Fig. 1
Fig. 1 Structure of the 2017 World Bank-supported CAT bond transaction (Source: Author's elaboration)

Fig. 2
Fig. 2 Evolution of the fiscal balance-to-GDP ratio for Mexico versus full sample and synthetic control unit, 2002-2016 (Source: Author's elaboration using R Core Team (2022)).a Comparison with full sample average, b Comparison with synthetic control unit

Fig. 3
Fig. 3 Fiscal balance-to-GDP ratio for Mexico and its synthetic control unit, 2002-2021 (Source: Author's elaboration using R Core Team (2022)).Note: Negative values on the vertical axis correspond to fiscal deficits, while positive values denote fiscal surpluses

•
(Andina 2019;Evans 2019ry of Finance of the Government of Peru has received a USD 60 million CAT bond payout in 2019.The payout was triggered by a magnitude 8.0 earthquake that struck the country's Loreto region on 26 May 2019, causing widespread damage to private homes, as well as to public buildings and infrastructure(Andina 2019;Evans 2019).• Philippines (2022): The Government of the Philippines received a USD 52.5 million payout in 2022 after Super Typhoon Odette (i.e., "Rai" in international jargon) struck the country on 16 December 2021 (DoF 2022; Evans 2022).The tropical cyclone impacted several regions in the Philippines, leaving more than 10.6 million people affected and causing severe damage to private homes, public infrastructure, crops, and agricultural facilities (ReliefWeb 2023).

Table 1
List of countries included in the donor poolSource: Author's elaboration based on data from EM-DAT (2023) and World Bank (2023a) Country Natural disaster type Natural disaster sub-type Short description of the natural disaster event ) compared to flow variables.Another important question is whether to use budget balance data for the general government versus the central government.I opt to use the former, as the general government has a wider coverage thus also capturing activities by non-central government entities that may play an important role in certain countries.For the reasons outlined above, the dependent variable in my paper will be Mexico's general government fiscal balance expressed as a percentage of GDP.

Table 2
Predictors of the fiscal balance used in the SCM

Table 3
Synthetic control unit weights for Mexico

Table 4
Pre-2017 characteristics of Mexico versus synthetic control unit and full sample mean