Assessing Socioeconomic Risks of Extreme Weather Events: Lessons from El Niño-Related Road Damages in Peru

Periodically, heavy rainfall associated with extreme El Niño events severely damage the Peruvian road-network. This study aims to assess the socio-economic effects of these non-frequent but recurrent climate shocks. For this purpose, we build a Computable General Equilibrium model that addresses the consequences of El Niño-related damages through an increase in interregional transportation costs and a negative externality effect on activities’ output, which has been estimated beforehand using a firm database. For designing simulation scenarios over a ten-year period where extreme El Niño events occurs at random as observed in the past in Peru, we model the occurrence of these events as a stochastic process with a VAR representation based on historical climatic data. Within this framework, we first show that such events constitute a significant one-off disaster risk for the country, threatening shifts of − 2.8% in GDP and +1.9% in poverty rates with an annual probability p = 1.4%. We further show that they also present a longer- term risk, leading to average annual deviations from normal trend by −0.8% in GDP and +0.4% in poverty rate with a probability p = 12.6% over a ten-year period. However, we finally show that Peru might reduce these risks in constructing more disaster-resilient road infrastructure.


Introduction
El Niño events present the most important adverse climatic threat to Peru, regularly causing major disasters in the country. For instance, in 1982-83, 1997-98 and 2017, heavy rainfalls associated with particularly severe events produced extensive landslides and flash flooding throughout the country, leading to widespread destructions of homes and agricultural productions as well as major disruptions of critical infrastructure such as water and sanitation systems and road networks. The direct economic losses of these damages are regularly reported.
In 1998 and 2017, they were for instance estimated to 3.5 billion USD (6.3% of GDP) and 9 billion USD (4.2 % of GDP), respectively [1,2]. However, their adverse higher-order cascading consequences for Peruvian economy and society remain surprisingly poorly investigated.
In this context, the main objectives of this study are to quantify the indirect socio-economic costs of extreme El Niño events in Peru and to assess the risk that such infrequent but recurrent shocks present to this country. Testing evaluation methods appears indeed to be of primary importance for assessing how these extreme weather events can be an impediment to Peruvian development and household welfare, and might help to design effective preventive strategies.
However, one key insight of the economic literature is that assessing the effects of such natural disasters is challenging (for comprehensive reviews, see e.g. Rose [3]; Okuyama [4]; Kelly [5]; Meyer et al. [6]; Jahn [7] or Botzen et al. [8]). On the one hand, the many interacting factors determining their ultimate indirect consequences in the short run as well as in the longer run necessitate dealing with complexity. On the other hand, the hazardous nature of such events and their low probabilities of occurrence also imply to deal with uncertainty.
For addressing both these issues, some methodological choices were therefore made upstream to our analysis. First, as many studies stress that addressing all impacts of natural disaster appears challenging (or even illusory) methodologically, we chose to focus on one main specific direct impacts of extreme El Niño events in Peru, which is their damages to road infrastructure. The Peruvian economy is indeed highly dependent on road transportation, which is the dominant mode for transporting people and products across the country. However, its road network relies on a few critical links, which are particularly vulnerable to heavy rainfall given the specific topography of the country. As an illustration, the 1997-98 El Niño event damaged 8.2% of national primary roads, while the 2017 event damaged 7.2% [1,2]. Second, we chose to use a Computable General Equilibrium (CGE) model for our analysis. In economic literature, a growing number of studies apply CGE modelling for a variety of extreme events and a variety of effects at the global, national, and local levels (see e.g. Rose and Liao [9]; Pauw et al. [10]; Koks et al. [11]; Mac Donald et al. [12]). One of its main benefits is indeed to convert the abstract Walrassian theory into an applicable simulation apparatus and therefore to represent a multi-market economy as a system of interdependent components in a fully coherent analytical micro-macro general equilibrium framework. With economic agents responding to relative price signals on both the supply and demand sides of the economy, it thus promises to shed light on a wide range of complex indirect effects of extreme events-related damages that would be ignored with other methodologies, such as empirical approaches (see e.g., Rodriguez-Oreggia [13]; Felbermayr and Gröschl [14]; Baez et al. [15]; Salazar et al. [16]) or input-output models (see e.g. Hallegate [17] ; Okuyama and Santos [18]). Third, in order to gain further in complexity, we chose to use a multiregional (MR) framework for our CGE model, rather than the standard economy-wide framework. Considering spatial dimensions appeared to be of primary importance for providing a deeper understanding of the effects of extreme El Niño event-related road damages in Peru. On the one side, this country has a structural "lagging regions" problem [19,20], with an important gap between its coastal regions and its other regions, that should be included in our general equilibrium analysis. On the other side, as observed in the past, this country is affected by two types of El Niño events with distinct regional distributions of damages [21]. Global events (like for instance in 1982-83 and 1997-98), whose effects are worldwide, affect mainly the northern coast and the northern Andean and Amazonian regions. Coastal events (like for instance in 2017), whose effects are more localised in Peru (and Ecuador), affect also the northern coast and, above all, the region of Lima, which is the economic core of the country. Fourth, we retain a dynamic framework for our MR-CGE model. A first benefit of this choice is to allow us to consider not only the one-off disaster effects of El Niño events for their year of occurrence but also their lasting effects, or even their cumulative effects if they occur several times over the period of simulation. A second benefit is that it also allows us to adopt a risk-assessment perspective for our analysis by including uncertainty features regarding the hazardous occurrence of El Niño events. Fifth, as CGE modelling results are only relevant at the representative household group level, we also linked our CGE model with a household survey-based microeconomic module that we use to capture the effects of road destructions in term of poverty.
Beyond the specific study case of Peru, the contribution of our analysis to the extant economic literature about the economic effects of extreme weather events appears finally twofold. The first contribution concerns the choices we made for considering how El Niñorelated physical road damages transfer to cascading indirect flow losses across the Peruvian economy in our general equilibrium modelling framework. As is commonly done in other CGE studies with a transport focus (see e.g. Shahrokhi Shahraki and Bachmann [22] for a review), we firstly consider rising transportation costs with associated ripple effects through increased prices for producers and consumers. However, and more originally, we secondly consider a negative externality effect on the productivity of Peruvian activities. Since the seminal works of Aschauer [23] or Munnel [24], there is indeed a broad consensus among economists that a proper functioning of core infrastructure such as road-network is closely related to overall economic performance, specifically in developing countries like Peru (see e.g. Romp and de Haan [25]; Straub [26] for surveys). To our knowledge, this potential adverse indirect effect of extreme events, which is by nature non-marketed and not related to inter-sectoral relationships, has never been included in other CGE studies. This is we seek to do in this work. However, as this effect plays a critical role in our simulations, we also chose to prior econometrically estimate its magnitude in Peru with a firm database at national level, instead of making arbitrary assumptions.
The second contribution of this study concerns the ways we define relevant scenarios to be simulated with the MR-CGE model and that are representative of the weather conditions usually experienced by Peru over a ten-year period. Following recent studies in climatic literature [27,28], we model the dynamic of occurrence of extreme El Niño events in the country as a stochastic process with a vector autoregressive econometric model based on historical climatic data. Then, applying a non-parametric bootstrap procedure, we generate numerous replications of ten-year series where Global and/or Coastal events occur at random. This article is structured as follows. Section 2 details the main features of the dynamic MR-CGE model. Section 3 explains the main hypotheses and procedures adopted in designing the scenarios for simulations and Section 4 presents the results of these simulations. Section 5 finally addresses the opportunity for Peru to implement some preventive actions to reduce El Niño event-related risks.

Dynamic MR-CGE model
Our CGE model features six regions for the Peruvian economy, one government agent, six groups of households (one per region), and nine non-transport activities plus one transport activity in each of the six regions (equations and variables are presented in supplementary material).
Its within-period specifications rely on standard assumptions of CGE analysis for a developing country (see e.g. Decaluwé et al. [29]). On the supply side, each producer of a sector j in a region r maximises its profit by combining skilled and unskilled labour with fixed capital with a standard Cobb-Douglas technology. On the income side, each agent receives factor revenues and transfer income from other agents. On the demand side, households' consumptions follows a linear expenditure system derived from maximisation behaviours of Stone-Geary utility functions. Government's consumption is supposed to be exogenous in each region. Intermediate consumptions are driven by fixed technical coefficients with Leontief specifications and demands for investment purposes are derived from nominal investments and distributed among commodities and regions in fixed shares. On the market side, each good can be sold in its region, in another region, or overseas. Constant Elasticity of Transformation functions govern the rates at which the proportions of these sales change according to the relative price variations, which are determined endogenously. Domestic goods of a specific region are assumed imperfectly substitutable with imported goods and those of other regions, given Armington specifications. Their prices are determined endogenously to equilibrate supply and demand in each regional market. For labour markets, the skilled and unskilled overall labour forces are fixed nationally, and workers flow freely across activities and regions with wage rate determined endogenously. In the capital market, we assume that all regions generate savings that go into a national pool to be allocated to investments at a national level. Finally, the nominal exchange rate is the numeraire for the economy and the current account balance is fixed.
The between-period specifications of the model are in a dynamic recursive framework.
They mainly concern the accumulation of capital in each activity at time t, which depends on an exogenous depreciation rate and on the volume of new capital installed in time t-1. In the public sector, the latter is exogenous. In private activities, it derives from an investment function and allocated in a putty-clay fashion across sectors and regions in accordance with returns to investment.
On these bases, for a given time t, the vector of prices and wages for which demand equals supply in all markets simultaneously defines the general equilibrium of the economy. Accordingly, the model can be used to generate time paths for endogenous variables over a given time-period by numerically simulating successive equilibriums under different scenarios of shocks. This period has been set relatively short (ten years) for excluding significant structural changes in the Peruvian economy and thus maintain the consistency of the model initial calibration. Given the purposes of this study, the first variables of interest for these 10-year simulations are the levels of road network infrastructure in each region r at time t ( , ).
In the spirit of Aschauer [23], they are supposed to determine the total factor productivity of each private activity j in each region r. In other CGE studies, this effect is often modelled through an indirect link between public expenditure on road infrastructure and productivity [30,31]. As shown in equations (1) and (2), we rather chose a direct link, modelled in a Hicks's neutral manner, through the variable , , in the production function and an externality parameter , . In the simulations, as shown in equation (3), we will thus suppose that , can be negatively affected (∆ ̅̅̅̅̅̅̅̅̅̅ ) by an extreme El Niño event for a time of occurrence t ( ñ , ̅̅̅̅̅̅̅̅̅ ). After an event, we will also suppose that exogenous specific public investments for recovery ( , ̅̅̅̅̅̅̅̅̅̅ ) progressively restore each damaged road. The second variables of interest for the simulations are the levels of transportation costs at time t for each product i that travels from its region of origin r to another region r′. These costs depend in part on the prices of the industry of each region that produces the transport service, which are determined endogenously in the model. However, they are also driven by interregional technical coefficients ( , ′ , , and , ′ , , ), which are determined by the distance (or travel time) between producers and buyers. As shown in equations (4) and (5), we will suppose that the occurrence of an extreme El Niño event increases these coefficients for domestic products and exported products (∆ , ′ , ̅̅̅̅̅̅̅̅̅ and ∆ , ̅̅̅̅̅̅̅ ) because of damaged road networks.

Data for model initialization and calibration
We obtain the initial general equilibrium of the economy from the last Social Accounting Matrix (SAM) available in Peru [32] that we earlier disaggregated between six regions. The Peruvian Institute of Statistics identifies three main regions in Peru under topographic, economic, and demographic criteria: the Costa (the Pacific coast area), the Selva (the Amazonian area), and the Sierra (the Andean area). However, in order to better consider the specific regional effects of extreme El Niño events on road networks, we further divided the Costa and Sierra territories between two transversal regions (North and South), and finally, given the weight of the region of Lima in the Peruvian economy, we also chose to isolate this specific area from the Costa. Table 1 indicates some descriptive statistics of each region based on this multi-regional SAM.  where , , , and , , , are respectively labour and capital (in monetary units), assumed to be exogenously determined by the wages and rental rates on labour and capital markets. , , is an urbanisation rate parameter (function of the population density of the region) at the firm's headquarters, which is included for consideration of the potential productivity gains from urban agglomeration [34]. Similar to Calderón et al. [35], , is the total length of the paved regional roads (in km) relative to the agricultural land (in square km). Regarding potential endogeneity concerns, we assume that the i-th firm's value added has a negligible effect on road infrastructure in our multilevel setup, and can be thus considered exogenous. For the cost of capital, we use the average interest rate in foreign currency for the Peruvian financial market as a natural instrument. For labour, given the time series nature of data, a second instrumental exploits autocorrelation information implied in a lagged indicator ( , , ′ −1 ) collected by the EEA survey along with the current year information. Table 2 presents the final estimate for , . The F-statistics provide strong evidence against the weakness of the instruments, and the externality parameters tend to exhibit significant results, ranging from 0.11 to 0.38 in the Costa and Lima regions, although it is non-significant for the Construction, Energy-Water, and Manufacture sectors. In the Sierra and Selva regions, they are all significant (except for the Energy-Water sector), ranging from 0.09 to 0.27.

TABLE 2
Finally, we also use the 2019-National household survey (ENAHO) in order to feed a microeconomic module linked with the MR-CGE model with a top-down approach [36]. In this perspective, consistent with the Stone-Geary utility function used for maximisation behaviours in the MR-CGE model, we assume Cobb-Douglas preferences for consumers in each region and thus derive individual equivalent consumption that are used to calculate poverty rates for each time t.

MAP 1
When it occurs, this warm phase of ENSO is able to change weather patterns worldwide. In the particular case of Peru, depending on its duration and intensity, it can lead to more or less intense rainfall. For instance, the 1982-83 Very Strong event was certainly the most negative shock experienced by the country. Starting during the agricultural period for 17 months, it contributed to a reduction of −7% in real GDP [38]. More recently, the consequences of the Very Strong 1997-98 event were estimated at −4.5% of real GDP [39]. However, things are a bit more complicated for Peru as, in addition to these Global events, another type of phenomena also drastically affects the country: the Coastal El Niño. This more localised event (unique to the Peruvian and Ecuadorian coastline) is characterised by a warming of SST along the Peruvian coasts but cooler conditions in other regions [21]. Since 2012, the Peruvian El Niño study agency (ENFEN) has used its own index (ICEN) in order to monitor such phenomena. It is based on SST anomalies exceeding +0.4 °C for at least three months in the easternmost Niño.1+2 Pacific region. Coastal El Niño events are often classified as weaker than Global events upon climatic criteria, mainly because of their shorter duration. However, they can produce comparable negative consequences in Peru based upon economic criteria. INDECI [2] estimated for instance that the 2017 event contributed to a −1.6% change of real GDP. Figure   1 summarises the occurrence of each of these different extreme El Niño events experienced by Peru since 1950.

Extreme El Niño event-related shock intensities
For considering the intensity of the shocks induced by each type of event in the scenarios for simulations with the MR-CGE model, we started by evaluating their respective physical damages on Peruvian road networks (∆ ̅̅̅̅̅̅̅̅̅̅ ). At the national level, the latter is close to 24500 km length, of which 70% is in asphalt and 30% is in more basic pavement [40]. As shown in  Once these damages have been located across regions, we secondly evaluated their associated increases of transportation costs across the country (∆ , ′ , ̅̅̅̅̅̅̅̅̅ and ∆ , ̅̅̅̅̅̅̅ ). For this purpose, using the index of link criticality proposed by Rozenberg et al. [41], we approximated the number of additional kilometres needed for transporting products after damage from an event. By comparing these additional kilometres to the total distance between regions, we therefore deduce the additional transport costs (in percent) associated with each road destruction. At this point, it should be stressed that we only assumed additional costs when a region experienced a major destruction rate (higher than 10%), and that the destination of exported products for all regions is always the region of Lima, given its weight (90%) in total exports for Peru.
Finally, our scenarios also include the levels of investments needed for road networks roads can be restored to their original design in four years. Accordingly, we will thus suppose in the simulations that, after a year of occurrence, an event has lasting (but progressively decreasing) effects on roads levels and transportations costs until the recovery process completes. Table 4 summarises the main hypotheses that we finally retained in our scenarios.

Extreme El Niño events' likely frequencies over a 10-year period
Given the hazardous nature of extreme El Niño events in Peru, the scenarios also consider the probabilities of occurrence of events over the 10-year period of simulation. In the climatic literature, physically-based models are the primary tool for evaluating these probabilities.
Nevertheless, although their ability to simulate El Niño-related dynamics at the global scale has improved markedly in recent years, they still exhibit errors in predictions regarding the amplitude, period, irregularity, or spatial patterns of the phenomenon [37]. In this context, alternative parsimonious approaches consist in using statistical estimates derived from past weather pattern analyses [27,28,43,44]. In line with these works, we thus modelled the likely annual occurrence of extreme El Niño events ( ñ , ̅̅̅̅̅̅̅̅̅ ) over ten years from the stochastic process that governs the SST anomalies. For this purpose, using NOAA data for SST over the 1870-2019 period in the four El Niño Pacific regions, we modelled the SST anomalies' crossdependence between the four regions as well as the self-dependence (autocorrelation) within regions by a VAR representation.
As a first step, we tested whether SST anomalies' series are stationary by means of an augmented Dickey-Fuller test with drift, where the optimal number of lags is chosen in such a way that the auxiliary regression residuals are white noise. Results in Table 5 show that the unit-root (null) hypothesis is rejected for the four series and that the stationarity can be thus accepted.

TABLE 5
As a second step, we therefore estimated a stationary VAR model of the four series in levels where the number of lags is chosen according to a Bayesian Information Criteria: is a 4 × 1 column vector of the monthly SST anomalies across the four regions at time t. is a 4 × 4 coefficient matrix that represents the cross-dependence and autoregressive behaviour of with respect to the j-th lag. The autoregressive order is given by . From this model, as described by Fresoli et al. [45], we implemented a non-parametric bootstrap procedure for ten years, which delivers a multi-array of 100 thousand replications of time series.
Every i-th replication is noted * , a matrix with 4 columns (the four Pacific regions) and 120 rows (the number of months over the 10-year period).

Simulation results and discussion
Following the standard procedure in dynamic CGE modelling, we first simulate the SCBAU scenario over the 10-year period by updating various constants and exogenous variables from one year to the next, using the annual population growth rate (which is close to 1.1%). Second, we identify among the 100 thousand time series ( * ) those that correspond to each of the six  Table 7 presents some selected results of these simulations.

Disaster effects of events for a year of occurrence
The first three columns in Table 6   Over the same time, public investments for recovery also produce ceteris paribus their own macroeconomic effects. First, by increasing the demand for the Construction activities that produce the capital goods required for restoring roads, they generate multiplier effects and therefore upward pressure on demand and prices on product markets. Second, given the savings-investment constraint driving the capital market's equilibrium, they also generate crowding-out effects for other investments for non-recovery purposes. All these combined effects finally contribute to a progressive decrease of the adverse consequences of events until the recovery process is completed.

Average effects of events over the 10-year period
These lasting effects of events, as well as their possible multiple occurrences over the period, are considered in a longer-term risk-assessment perspective in the next columns in Table   7. The SCN scenario corresponds to the occurrence of at least one extreme event (either Global In the most unfavourable (however unlikely) scenario (SCN =3 with 3 = 0.04%), if three events occur over the 10-year period they cause average annual deviations of −1.74% in GDP, 5.16% in prices, −1.77% in income per capita and +1.11% in the poverty rate.

Exploring options for reducing the extreme El Niño event risk in Peru
In previous simulations, the socio-economic risk associated with extreme El Niño events depended on the occurrence of hazards on the one hand and on the level of vulnerability of regional road networks on the other. If reducing hazard occurrence is not an option, preparing appropriate preventive plans that allow the improvement of the road network's resilience to event impacts might be of great interest. Currently, such a perspective is not really considered by Peruvian authorities as, for past events, planning and preparation were only characterised by 'emergency' short-term ex-ante measures conducted in the months immediately prior to events' predicted onset or by ex-post interventions for recovery [46]. Yet, there seems to be some room to manoeuvre in constructing more disaster-resilient road infrastructure and to accommodate the heavy rainfall associated with these events. As shown by the Road Quality Indicator [47], the Peruvian road network is indeed in a particularly poor condition, only ranking 111th of 141 countries in 2019. In this context, as suggested by the World Bank [40], a first action could consist in progressively upgrading the most fragile primary roads (that is those destroyed during the 1997-98 and 2017 events) to flood-and landslide-proof status. Adding tunnels, raising the level of roads, and strengthening structures and drainage capacity could help to achieve this. A second action could consist in increasing the redundancy of these critical links by upgrading existing alternative second-best routes to primary paved roads so as to be equivalent to firstbest roads. In a first approach, we consider these options to be implemented at national level over the 10-year period. However, in a second, more pragmatic approach, we consider that only the two regions most affected by El Niño events, that is Costa Norte and Lima, can benefit from such investments. The respective costs of these plans are taken from the World Bank [40]. Table 8 presents selected results of the simulations of previous stochastic scenarios when Peruvian government implemented such preventive plans. They show that all strategies might appear worthwhile in buffering the expected disaster effects of events as well as their negative average longer-term effects over the 10-year period, without however fully eliminating the risks. In our general equilibrium framework, they also show that targeting the most impacted regions should be preferred to nation-wide plans with strong crowding-out effects on other investments for non-infrastructural purposes. Similarly, given its positive externality effects on activity productivity, extending the current primary road network by upgrading extant secondbest roads should be preferred to upgrading the most fragile primary regional roads.

TABLE 8
To gain further insight into these results, Table 9 finally presents two cost-effectiveness indicators that allow us to address the critical question of how much each million USD annually invested over 10 years actually minimises the consequences of extreme El Niño events on national GDP and poverty rate. Regarding these criteria, results show that investments might generate positive average net returns in terms of growth and poverty, if at least one event occurs over the period ( = 12.59%). However, when considering the very likely possibility that no event occurs, the only profitable strategy would be to upgrade the extant second-best roads in the Costa Norte and Lima.

Conclusion
Using a multiregional dynamic CGE model and a historical-based bootstrap procedure for designing relevant scenarios of the occurrence of extreme El Niño events in Peru over a 10- year period, this study shows that their related road damages constitute not only a significant disaster risk for this country, but also a longer-term risk threatening its economic performance and household welfare. However, it also shows that some profitable options might be implemented for managing this risk, such as, for instance, increasing the redundancy of the most fragile critical roads in the most affected regions. If these findings offer a valuable indication of extreme El Niño event risk levels for Peru and a useful guideline for conducting preventive actions, some caution should however be exercised when interpreting their absolute magnitude, as some caveats must be considered in our methodology.
First, this study does not fully address the challenge of complexity of assessing extreme El Niño event indirect effects on the Peruvian economy and therefore underestimates their real related risk. By focusing solely on road damages, it ignores all other effects that clearly alter economic performances and household welfare. For instance, the impacts on the agricultural sector could also be considered, given the numerous forward and backward linkages of this sector with other sectors and its dominant influence on the incomes of the poorest populations.
However, such inclusion is not straightforward, as it would require better adapting the modelling framework to accommodate the very specific conditions prevailing under a disaster situation. In this spirit, one promising line of research could be to include temporary disequilibria assumptions in the MR-CGE model arising from supply bottlenecks, shorter-run behaviour for economic agents, or restrictions on the movements of labour.
Second, this study does not also fully address the challenge of uncertainty in evaluating extreme El Niño event effects and/or occurrence. On the one side, regarding the valuation of events' respective damages to Peruvian roads, we had no choice but to refer to the 1997-98 and 2017 events. However, as suggested by French and Mechler [46], some variability could be expected between the impacts of events of a same type. On this front, the United Nations Office for Disaster Risk Reduction has recently started to develop international time-series data (including for road damages) that could help assist in better addressing risks in the future. On the other side, regarding the frequencies of events in Peru over ten years, the potential influence of a warming climate on ENSO variability is a source of uncertainty that we did not consider in our analysis. Although there is still no clear consensus regarding an intensification of El Niño events in the 21st century, some studies however confirm a possible risk of more frequent stronger episodes, especially for Coastal events [48]. However, by imposing a constant volatility of SST anomalies in the VAR modelling, our historical-based bootstrap procedure cannot take into account this risk. An alternative approach, where SST anomalies dispersion is modelled by means of a conditional heteroskedasticity framework, could certainly usefully be explored in order to better consider this potential recent (and future) influence of climate change. Source: Own calculations from the regional Social Accounting Matrix for Peru  Notes: (1) Bootstrap standard errors clustered at the regional level *p < 0.05, **p < 0.01, ***p < 0.001 ; (2) No estimates for Oil-Mining in the Sierra Norte, Sierra Sur and Selva due to the very low degrees of freedom resulting from the few observations in the database       Notes: (1) GDP-Benefit-cost ratio is the net gain (in Million USD) in real GDP over the period for one Million USD invested annually;

LIST OF TABLES
(2) Poverty-Benefit-cost ratio is the absolute reduction of the number of poor over the period for one Million USD invested annually.     , ,0