Multiple Hazards and Economic Resilience: Sectoral Impacts and Post-Disaster Recovery in a High-Risk Brazilian State

Abstract

Rio Grande do Sul accounts for 22% of Brazil’s losses from extreme events, mainly droughts and floods. The state had the second-worst economic performance in the country between 2000 and 2022. This study quantifies the impacts of major events such as droughts, floods, and the COVID-19 pandemic on economic sectors. Three methods were applied: structural breaks, recovery time, and sector-specific loss estimates. The analysis covers 15,365,123 observations of monthly invoice values from January 2017 to April 2025, involving 357,001 companies paying value-added tax on consumption. The results indicate that negative structural breaks occurred in a few sectors, which account for 5% of the state’s economy. The recovery time followed a similar trajectory between droughts and COVID-19. On average, sectors took 12 months to recover from COVID-19, compared with about 6 months for natural hazards. The sectors most impacted were travel, artistic activities, machinery and equipment industry, accommodation, and domestic services. Aggregated loss estimates were highest during the COVID-19 pandemic (−8%), followed by floods (−1%) and droughts (0%). The results indicate remarkable overall short-run economic resilience. Furthermore, sectors such as information technology, consulting, business services, and healthcare performed exceptionally well.

1. Introduction

Rio Grande do Sul (RS), the southernmost state in Brazil, has faced recurring extreme weather events such as droughts and floods, which intensified after the COVID-19 pandemic. Although the economy of RS is relatively diversified, agriculture and agribusiness continue to play a fundamental role in its overall production. Between 2000 and 2024, however, RS recorded the second-lowest Gross Domestic Product (GDP) growth rate among all Brazilian states. It also recorded the highest economic losses from weather events, accounting for 22% of total damage reported across public and private sectors nationwide. Following RS are Minas Gerais (MG) with 10%, Santa Catarina (SC) with 9%, and Paraná (PR) with 8% (Figure 1). The southern region of Brazil stands out as a hotspot for extreme events, representing 39% of the country’s total losses.

Figure 2 shows that RS lost over BRL 140 billion between 2000 and 2024, more than double the losses recorded in Minas Gerais and Santa Catarina. In terms of the number of events, Minas Gerais reached 8635, Rio Grande do Sul 8369, and Santa Catarina 7594. Bahia is the fourth state with the highest number of events with 4952.

An analysis of the 2002–2022 period reveals that Rio Grande do Sul (RS) showed relatively weak economic performance. RS recorded an average growth of just 1.50%, ranking only slightly ahead of Rio de Janeiro (RJ), which posted the lowest growth rate in the country at 1.47% (Figure 3). In contrast, Mato Grosso grew at an average of 4.9% during the same period, nearly doubling its share of the national GDP. Meanwhile, the states of Sergipe, São Paulo, Rio Grande do Sul, the Federal District, and Rio de Janeiro saw notable declines in their respective contributions to Brazil’s GDP.

The state that is the subject of this study, Rio Grande do Sul, is characterized by a strong presence in the primary sector, particularly in agriculture and livestock, and plays a significant role in Brazil’s export economy.

Table 1. Main 18 economic sectors in RS, average percentage in NF value from 2019 to 2024, and number of companies in 2024.

Code	Sector	Average Perc. 2019–2024	Number of Companies 2024
			Standard	Simplified Tax Regime	Total
46	Wholesale trade, except motor vehicles and motorcycles	26.52	7617	7322	14,939
47	Retail trade	20.68	23,559	73,093	96,652
10	Manufacture of food products	10.97	1809	3129	4938
45	Trade and repair of motor vehicles and motorcycles	5.15	4107	12,711	16,818
20	Manufacture of chemical products	4.95	478	356	834
28	Manufacture of machinery and equipment	4.69	1037	1129	2166
19	Manufacture of coke, petroleum products, and biofuels	4.09	23	4	27
29	Manufacture of motor vehicles, trailers, and bodies	3.66	283	337	620
25	Manufacture of metal products, exc. machinery, equipment	2.45	923	3722	4645
22	Manufacture of rubber and plastic products	1.85	661	816	1477
15	Preparation and manufacture of leather goods and footwear	1.65	505	629	1134
12	Manufacture of tobacco products	1.04	24	27	51
56	Food	1.07	2012	14,573	16,585
17	Manufacture of pulp, paper, and paper products	1.15	199	229	428
11	Manufacture of beverages	1.18	253	581	834
31	Manufacture of furniture	1.09	423	2231	2654
24	Metallurgy	1.18	104	143	247
35	Electricity, gas, and other utilities	1.06	167	0	167
	Total	94.41	44,184	121,032	165,216

presents the main economic sectors in the state, measured by the value of invoices subject to the state’s consumption tax. The commerce sector ranks first, followed by food manufacturing. As of 2024, the state had approximately 165,000 companies, with around three-quarters classified as small businesses operating under a simplified tax regime. In terms of sectoral distribution, commerce stands out with nearly 100,000 companies. Notably, there is also substantial activity in vehicle and motorcycle-related commerce, followed by the food industry and the wholesale sector.

Table 2. Types of events that occurred in RS between 2000 and 2024, sum of updated value, and number of incidences.

Event	Sum Loss Public and Private	Frequency
Drought and dry spell	99,991,029,811	3272
Heavy rains	16,877,598,521	1195
Flash floods	8,111,654,401	1408
Floods	5,866,260,674	529
Windstorms and cyclones	5,354,169,047	1080
Hail	3,809,812,126	654
Flooding	269,355,708	103
Cold wave	187,287,720	18
Tornado	160,653,724	31
Others	152,135,881	22
Mass movement	59,752,093	43
Erosion	10,999,763	2
Forest fire	6,816,583	7
Dam failure	1,746,494	2
Infectious diseases	17,797	2
Heat wave and low humidity	3420	1
Total	140,859,293,762	8369

National Secretariat for Civil Defense and Protection—SEDEC/MIDR—and the Center of Studies and Research in Engineering and Civil Defense—CEPED/UFSC.

presents the total public and private losses in Rio Grande do Sul (RS) since 2000—adjusted for inflation—alongside the frequency of hydrometeorological events recorded at the municipal level. Droughts emerge as the most frequent and economically damaging phenomenon, although heavy rains and floods also occur with high regularity. RS continues to face significant exposure to a range of hydrometeorological hazards. According to the Governor’s 2024 message to the Legislative Assembly, droughts are the most common type of disaster in the state, followed by floods [1]. A study examining the duration of droughts in 2020, 2022, and 2023—based on the time required for economic recovery at the municipal level—found that, following each drought event, at least 75% of the affected municipalities recovered economically within three months, while 24% of municipalities were unaffected by any of these events [2].

It is important to note that according to the Inter-American Development Bank (IDB) and the World Bank, the total effects of floods in RS in 2024 were estimated at BRL 88.9 billion, using their own methodology, and this was the largest catastrophe in the history of RS. The flood directly affected 876,565 people, with 95 municipalities declaring a state of calamity and 323 declaring a state of emergency. A study showed that the municipalities of RS have limited fiscal response capacity for these events [3].

The objective of this study is to estimate the impacts of extreme events that occurred between 2020 and 2024 on the economic sectors of Rio Grande do Sul (RS) and to identify those sectors most susceptible to crises. The analysis includes estimates of both the magnitude of losses and the duration of the impacts. The methodology combines the Bai and Perron (1998) structural break model [4], Kaplan–Meier survival analysis adapted to measure recovery time, and sector-specific loss estimates across 74 economic sectors. The data employed consist of the total monthly invoice values issued by companies operating in the state, covering both B2C and B2B transactions. Each company is classified by sector according to the Brazilian Institute of Geography and Statistics (IBGE).

Following this introduction, Section 2 provides a review of the literature on the economic impacts of natural disasters. Section 3 describes the methodologies used, including descriptive statistics of the dataset and a detailed explanation of the procedures adopted. Section 4 presents and discusses the results, and finally, Section 5 summarizes the main conclusions of the study

2. Literature Review on Economic Impacts of Natural Disasters

The 2015 Paris Agreement emerged as the most important global scientific safeguard for climate action. The quest to limit the increase in global average temperature to 2 °C above pre-industrial levels and pursue a reduction to 1.5 °C will greatly reduce the effects of climate change [5]. A study on the effects of climate change, considering the non-linear impact of historical temperature, shows that rich and poor countries behave similarly, indicating that wealth does not guarantee lower vulnerability. It indicates that failure to mitigate global warming can reshape the global economy and increase inequality [6]. The demand for research, information, and action to address the challenges of a changing climate continues to grow [7]. The damages have been noted in several countries [8,9,10,11].

To better analyze climate risks for economic agents, beyond generic and global or national forecasts, it is necessary to map and quantify them. A study of the Province of Belluno (Italy) was conducted for its main sectors: summer tourism, winter sports and events, the eyewear industry, and electricity supply; with stakeholder involvement, the researchers provided socioeconomic agents with clear messages about how their activities could suffer or benefit from climate change [12]. The study analyzed the propagation of sectoral shocks in the economy. The analysis began with the impact on one sector and then tracked spillovers to others, showing that supply and demand shocks strongly affect both sectoral gross value added and a sector’s share of the economy. Regarding the pandemic specifically, it pointed out that the spillover effects are considerable, making up a significant part of the reduction in economic activity in 2020 [13].

A study on the medium- to long-term effects of pandemics in relation to other economic disasters points to significant and persistent macroeconomic effects, with substantially reduced real rates of return. The capital stock does not decrease as it does in wars, but the pandemic can lead to a relative shortage of labor and also to a change in the savings pattern [14]. Researchers found that after a pandemic, output is interrupted for about seven years, although the duration varies across countries and economic sectors. Pandemic shocks lead to a short-term drop in the number of patent applications. Effects on growth are also expected [15]. A study conducted in Greece examined the impact of economic shocks on the resilience and recovery of administrative regions. The findings indicate that regions heavily reliant on tourism shifted from high to low resilience during the COVID-19 pandemic. Strong local economic activity was identified as a key factor supporting resilience, whereas nationally dynamic sectors with external market dependencies proved more vulnerable and slower to recover. The study highlights the importance of strengthening local economic structures and diversifying activities as essential strategies for ensuring long-term sustainability [16].

A study conducted in Brazil highlights the significant impact of the COVID-19 pandemic on small businesses, caused by the sharp drop in demand and interruption of activities. In terms of capital stocks, of micro and small lost an estimated BRL 9.1 billion to BRL 24.1 billion by June 2020, mainly in the trade and services sectors. It is estimated that it will take 1 to 3 years to recover this capital, depending on government support to reduce this period [17]. Another study focused on small businesses analyzed the survival of 8931 firms from 2017 to 2023 in Rio Grande do Sul, Brazil. The results indicate that survival rates were lower in the commercial sector and in financial intermediation activities. When detailing the commercial sector, which accounts for the vast majority of companies, the sectors that suffered the most were retail, accommodation, and food in terms of survival rates. The results indicate that small businesses remained relatively resilient during the COVID-19 pandemic, signaling the effective government support. The smallest companies with revenues below USD 15,576 per year were the most affected, with only 39% survival after 7 years [18].

Looking at territorial scales and the possible impacts of climate change on the Brazilian economy in terms of temperature and precipitation, a study indicates that climate change will increase the concentration of activity in space and reinforce regional and social inequalities, with a reduction in well-being in rural areas and the potential to create pressure on urban agglomerations. However, some sectors and regions are likely to benefit from the process [19]. Brazilian grain production declined in 2021 after several years of growth, primarily due to reduced agricultural productivity caused by a lack of rain, frost, and low temperatures in the country’s main producing regions. Nationally, this drop in productivity led to a GDP reduction of approximately 0.30% in 2021, impacting household consumption, investment, exports, employment, and capital stock. These effects are expected to have long-term implications, with an estimated GDP reduction of 0.14% projected for 2035. The most affected states were Mato Grosso do Sul, Paraná, and Amazonas, while Rio Grande do Sul was among the least affected in that particular year. The sectors that were hit the hardest were corn production and civil construction [20].

The economic feasibility of adopting climate change adaptation measures in Italy’s agricultural sector is supported by a study that identified several barriers to their adoption, including the difficulty of accessing knowledge about good practices and the lack of resources for investments. The economic feasibility was calculated based on the potential reduction in climate-related damages [21].

The study investigated multiple disasters and their economic effects. It included floods, pandemic control, and export restrictions and considered inter-regional substitution and production specialization influencing the economy’s resilience to events. The results suggest that a strict but short-term pandemic control policy would reduce economic costs when multiple crises occur simultaneously [22]. A study in China on worker productivity due to increased heat stress and the impact on economic sectors identifies that workers in outdoor environments will be seriously affected, while indoor workers may benefit from the massive use of air conditioning units. However, all regions will face the negative economic impacts of increased heat stress due to climate change. Agriculture and construction will suffer the most severe economic impact, but the effects on the manufacturing and service sectors should not be ignored due to their magnitude on GDP. At the national level, long-term GDP losses may remain limited to 1% in the most optimistic scenario [23].

The drought and bushfires in Australia have reduced its well-being. In addition to the direct losses, there are also effects on livestock production and investment. There is potential future economic loss to the international tourism sector from the impact of the bushfires, which is the subject of debate [24]. Beyond short-term effects and damage to physical capital, it is necessary to analyze the long-term effects of natural disasters. Events can promote increases in human capital, contributing to a certain compensation, and can also stimulate the development of new technologies, even contributing to an increase in total productivity [25]. Studies emphasize that decentralized governments in terms of fiscal structure suffer fewer fatalities during extreme events, but they need a sufficient level of human capital [26]. After a disaster, companies often increase inventories, motivated by possible supply chain disruptions and, more significantly, by shifts in risk perception [27]. This tends to speed up recovery, due to an unexpected effect.

To face competition, strategies can be summarized in three aspects: cost leadership, differentiation or segmentation, and reach. Not all sectors are equally attractive, but even less competitive firms can remain financially sustainable if the sector has high profitability [28]. An important role is played by barriers to entry in certain sectors, which, in their absence, can be a factor of strong competition, although not all entrants have the same initial conditions. The entrant can be a newcomer or an established company that wishes to diversify [29]. Commerce, for example, is a very competitive sector, with thousands of companies of all sizes.

3. Methodology

Three methods were used to understand the impact of events on economic sectors. First, we used the Bai and Perron method to detect structural breaks. The structural breaks were analyzed based on total sales adjusted for Brazilian inflation (IPCA) as of April 2025. Twelve-month moving averages were unsuitable because they produced an abnormally high number of breaks due to their relative stability. This first method emphasizes major crises, capturing structural breaks involving shifts in both intercepts and trends.

If the 12-month moving averages were used instead of the monthly values corrected by the IPCA (real values), there would be an overdetection of breaks. This occurs for three main reasons. The first is the delay and smoothing of the moving averages, generating slower responses to shocks. This may seem like it should reduce the number of breaks, but in reality, the opposite happens, since the moving average carries the effects of a shock for several months. This generates smooth ramps and non-linear trends, and the Bai and Perron method, when adjusting for trend breaks or changes in levels, detects multiple break points when trying to follow these smooth curves. Another point is that the moving average generates gradual transitions that violate the assumption of stable segments, and the model tries to compensate with several artificial breaks, even if the original series contains only a single shock. Another point is that the reduction in noise generates an increase in the detection of secondary trends that can be interpreted by the algorithm as structural changes.

In the second analysis, we examined how long sectors took to recover from crises. We calculated recovery times for three events: COVID-19 in 2020, droughts in 2023, and floods in 2024. This model, based on Kaplan–Meier survival analysis, identifies the probability of recovery. From the generated base, we continued an analysis of the time until returning to the pre-crisis situation by sector. The threshold was defined as a 15% drop in monthly company sales, corresponding to a 1.25% decline in the 12-month moving average. Finally, we consolidated the results of the companies using the average of each sector. The analysis included 76 sectors with complete data across 76 months (January 2019–April 2025), forming a time series with 5776 observations.

Survival analysis (SA) is a widely used technique, especially the Kaplan–Meier estimator [30] and the Cox model [31]. The response variable for this type of analysis is the time elapsed until the occurrence of an event of interest, which is called the failure time. In this article, the event under study is the recovery from the economic crisis of the economic sectors. Thus, the occurrence of the event is a positive factor—a recovery—unlike the traditional model, which represents death. SA aims to determine the probability of survival and the risk of an event for a group of individuals based on the time itself and variables called covariates [32]. The analysis includes censoring—its main characteristic— since some economic sectors did not experience a relevant downturn during these events [33]. Censorship functions as a control group. Sectors that did not recover until a certain time are also censored. Observations of each of the events were censored according to their characteristics, with COVID-19 being 24 months and Droughts and Floods 12 months. The effects of COVID-19 lasted an average of 5.9 months—nearly twice the duration of droughts (3 months) and floods (3.8 months). Consequently, the threshold for COVID-19 was set at twice that of other events.

In the third analysis, the sectoral losses or gains from 2020 to 2024 were estimated. This required two more years of analysis with data since January 2017. We used 72 months of training for the models. In this analysis, it was only possible to use 74 sectors. We used sales values grouped by sector on a monthly basis. The training model used 72 months, that is, from January 2017 to December 2022, and the test model used 28 months ahead. Linear, fourth-order polynomial, Prophet, and neural network autoregression (NNAR) models were applied. To determine the best model for each sector, we compared the Root Mean Squared Error (RMSE)—the mean squared difference between actual and estimated values—and chose the model with the lowest error for the entire period. Figure 4 summarizes the methods used.

Evaluations using appropriate tools allow us to verify the real effects of policies [34]. In some situations, in economic life, changes can be easily detected, but in others, statistical tests are necessary to distinguish between conditions at one time and another. There are cases in which conventional tests do not detect changes because there is a structural change in which the values of the model parameters do not remain consistent over time [35].

Many studies on structural change or breaks focus only on single break cases. In this study, we use a method that identifies multiple structural changes, which occur at unknown dates, using a linear regression model estimated by least squares. The main advantages are related to the properties of the estimators, the estimation of break dates, and the construction of tests that allow inference about the presence and number of structural changes and breaks [4]. Forecasting analyses that assume a constant and time-invariant data generation process, implicitly ruling out structural changes, may neglect significant aspects of the real world and provide inaccurate forecasts. Forecasting models that include the possibility of structural disruptions may increase their effectiveness and resilience [36].

In this study, the exogenous information was also incorporated indicating probable dates of disruption. If there is prior information, exogenous to the data, suggesting a potential disruption on a given date due to an institutional change, this information can and should be used without restriction. Exogenous information does not distort the model’s properties of the dimension and will likely produce a more powerful test. However, it is common for the data to indicate a slightly different date. This discrepancy may arise because major adjustments or changes in behavior often do not coincide precisely with the date of the institutional change [37].

We incorporated the periodization of COVID-19 waves in Brazil as exogenous information. The study identified three distinct waves: the first from 23 February to 25 July 2020; the second from 8 November 2020 to 10 April 2021; and the third from 26 December 2021 to 21 May 2022 [38]. According to the records of the Ministry of Integration and Regional Development (MIDR), we also adopted the periods of drought and flooding, with floods occurring in April, May, and June of 2024 and droughts in January, February, and March of 2023. Thus, the structural breaks were expected to occur around these periods.

The analysis of COVID-19 has been the subject of several time series forecasting methods, from the autoregressive moving average model (ARIMA) to the neural network autoregression model (NNAR), as well as hybrid models, which have been shown to facilitate decision-making by authorities [39]. Many studies have analyzed forecasting methods. In stock price forecasting, for example, the use of the Linear Regression, Polynomial Regression, and the Autoregressive Integrated Moving Average (ARIMA) model reveals their respective strengths and limitations. Linear regression provides a general overview of trends, while polynomial regression offers deeper insight into price variations and cyclical trends. The ARIMA model demonstrates superior short-term accuracy. Forecasting models are often extremely complex [40].

In summary, we can say that linear regression is simple and easy to interpret, polynomial regression captures non-linearities, the autoregressive neural network models non-linear patterns in residuals or directly in the series, and Prophet models seasonality and specific effects (e.g., festivities, holidays) through regression by segments. In this study, we used the monthly sum of invoices issued by companies to consumers and between companies. Through these data, we have a proxy for the level of economic activity across sectors. The initial database contained 15,365,123 observations, with 357,001 companies, of which 3.9 million observations were from companies in the general tax regime and 11.4 million pieces of information from companies in the simplified tax regime.

Table 3. Descriptive statistics of variables.

Variable	Description	Min	Median	Mean	Max
year	Year	2017	2021	2021	2025
month	Month	1	6	6.39	12
cod_firm	Identification of the company	0	17	58	979
sector	Sector national code	0	47	42.93	97
category	Tax regime	regular	3,933,169	simplified	11,431,954
sai_nfe_nfc	Monthly value of sales invoices	0	35,000	504,300	5,285,000,000
IPCA	Inflation index	1	1.288	1.268	1.518

describes the variables.

4. Results and Discussion

The first analysis using the structural breakdown method allowed us to identify the number of breakdowns and in which sectors and events. The second analysis focused on recovery time and the last on the estimated annual losses by sector.

4.1. Results of Structural Breaks

Structural breaks marked by declines in economic activity were identified during COVID-19, the 2023 droughts, and the 2024 floods in Rio Grande do Sul. A total of 82 breaks were observed across 20 sectors, representing 4.78% of economic activity, based on the annual average of 2024. Sectors’ structural breaks accounted for about 95% of economic activity (

Table 4. Sectors with structural breaks.

Code	Sector	Breaks	Sector %
26	Manufacture of computer, electronic, and optical equipment	1	0.60%
27	Manufacture of electrical machinery, equipment, and materials	2	0.58%
23	Manufacture of non-metallic mineral products	1	0.56%
16	Manufacture of wood products	2	0.53%
49	Land transport	3	0.45%
32	Manufacture of various products	2	0.41%
14	Manufacture of clothing and accessories	1	0.29%
13	Manufacture of textile products	2	0.26%
8	Extraction of non-metallic minerals	2	0.15%
33	Maintenance, repair, and installation of machinery and equipment	3	0.14%
52	Storage and auxiliary activities for transport	2	0.12%
38	Collection, treatment, and disposal of waste	2	0.12%
18	Printing and reproduction of recordings	2	0.09%
21	Manufacture of pharmaceutical and pharmaceutical products	1	0.08%
61	Telecommunications	2	0.07%
43	Specialized services for construction	1	0.05%
5	Coal extraction	2	0.04%
58	Publishing and integrated publishing with printing	2	0.04%
42	Infrastructure works	1	0.03%
95	Repair and maintenance of computer equipment	2	0.02%
	20 sectors	36	4.64%
	Other 26 sectors	46	0.14%
Total Breaks		82	4.78%
No breaks			95.22%

). It is important to note that some sectors comprise a large number of companies and are highly competitive, which facilitates the expansion of some firms while others contract. Similarly, less affected municipalities often compensate for disruptions in more affected areas. Detecting structural breaks at the sectoral level is far more complex than identifying them in individual companies.

Of the 82 structural failures identified, we analyzed their causes by event (

Table 5. Structural breaks by event identification.

Event	2019	2020	2021	2022	2023	2024	Total
COVID 1st Onset	0	14	0	0	0	0	14
COVID 2nd Onset	0	3	8	0	0	0	11
COVID 3rd Onset	0	0	3	0	0	0	3
COVID 3rd Onset–Drought 2022	0	0	0	3	0	0	3
Drought 2023	0	0	0	0	3	0	3
Floods	0	0	0	0	0	8	8
Not detected	4	3	11	11	10	1	40
Total	4	20	22	14	13	9	82

). About half of these failures occurred outside the suggested periods, likely due to delayed effects or other contributing factors. COVID-19, particularly the first wave, was the leading cause of structural failures. The 2024 floods ranked second, also contributing significantly to the total number of failures. It is important to note that a structural failure refers to a substantial disruption within an economic sector. May was the month with the highest number of failures, totaling 13—of which 8 were attributed to the 2024 floods. February recorded the second highest number, with 11 failures.

4.2. Recovery Time Results

The second analysis on recovery time calculated how many months companies needed to recover from the crisis. To calculate the crisis, the 12-month moving average was used, and a composite index based on the previous average of 0.9875, which is equivalent to a 15% drop in the month of the crisis, was converted into a moving average value of 1.25%. The time windows for verifying the state of crisis followed the following sequence: COVID-19, time = 14 to 19; droughts, time = 49 to 51; floods, time = 64 to 66. Time starts in January 2019, so COVID-19 begins to be detected in February 2020, for example. The censoring times were set at 24 months for COVID-19 and 12 months for climate events. Therefore, companies that had not yet recovered at that time were censored and assigned the maximum value.

Figure 5 shows the proportion of companies that had not yet recovered up to each point in time. In month 12, for example, 69% of the companies affected by floods had not yet recovered, with 55% showing no signs of crisis. In other words, 31% of the companies had already recovered, and in total, 86% were in normal conditions. After one year, about 15% of companies had not recovered from COVID-19, and 14% had not recovered from climate and hydrological events. It is possible to identify a similarity between the impact of COVID-19 and droughts in the first 12 periods, where we have up to 82% probability of non-recovery. In the case of floods, we see the probability of non-recovery is lower in the first month, and in month 12, the probability of non-recovery is still 12 points lower than in the case of droughts and COVID-19. This difference reflects the floods’ more localized impact: many areas declared a state of emergency, but not a state of calamity. The traditional Kaplan–Meier model estimates the survival function, that is, the probability that the company has not died by a given time. In this case, the model is adapted so that the event is the recovery of the company rather than its death, so the curve shows the proportion that has not yet recovered by each point in time (Figure 5).

Many companies did not experience a crisis during these events, or at least not at the magnitude we used as a threshold. A total of 65% of companies had already recovered or did not feel the effects within the first month in the case of COVID-19 or droughts, while only 55% did so in the case of floods. After 12 months, we can say that approximately 15% of companies are still below their pre-crisis level. This suggests a sharp recovery within 3–4 months, reaching 80%, followed by a slower, more gradual recovery.

Table 6. Main sectors by disaster, percentage of firms in crisis, and average recovery time (months).

Sector	COVID-19		Droughts		Floods
	Percentage of Firms in Crisis	Months	Percentage of Firms in Crisis	Months	Percentage of Firms in Crisis	Months
Retail trade	33%	12.05	32%	7.30	44%	6.76
Trade and repair of motor vehicles	36%	10.69	34%	5.87	45%	5.45
Food	30%	14.62	21%	6.27	38%	7.03
Wholesale trade	35%	10.03	35%	7.03	48%	6.39
Manufacture of metal products	41%	8.66	44%	6.40	52%	5.80
Food manufacturing	37%	11.43	30%	6.96	44%	6.95
Manufacture of furniture	43%	9.22	47%	6.32	52%	5.07
Manufacture of machines and equipment	50%	7.91	52%	6.54	60%	5.76
Wood manufacturing	46%	9.01	49%	6.46	58%	5.79
Clothing and accessories	45%	11.37	39%	5.79	56%	6.51
Land transport	21%	9.78	26%	6.13	36%	5.64
Average	34%	11.47	33%	6.88	45%	6.43

analyzes the key sectors of Rio Grande do Sul’s economy by month, indicating the percentage of companies affected by each event and the average recovery time for each crisis. During COVID-19, the food and commerce sectors took the longest to recovery—14 and 12 months, respectively. In the case of droughts, commerce and wholesale were the most affected, though the impact across sectors was relatively uniform, with an average recovery time of 6.8 months. The flood events had the highest percentage of companies in crisis, reaching 45%, compared to 34% during COVID-19 and 33% during the droughts. Companies in the food sector again took the longest to recover, averaging 7 months. Notably, during the floods, the variation in recovery time across major sectors was minimal. Overall, droughts and floods exhibited similar average crisis durations across various sectors.

4.3. Loss Estimates

Loss estimates by economic sector were calculated across 74 sectors using four different models. The Root Mean Squared Error (RMSE)—which measures the mean squared difference between actual and estimated values—was used to evaluate model performance. The model with the lowest RMSE for the entire analysis period was selected. The neural network model was the most frequently selected, chosen 38 times, followed by linear regression, which was chosen 20 times (see

Table 7. Models and best adjustment.

Model	Best Fit	Description
Linear	20	Simple linear regression of Sales as a function of time
NNAR	38	Autoregressive neural network for time series
Polynomial	3	4th-degree polynomial regression for complex curvatures
Prophet	13	Used for time series, with annual seasonality
Total of sectors	74

).

Table 8. Resume of number of sectors with losses from 2020 to 2024.

Item	2020	2021	2022	2023	2024
Number of sectors with losses	54	24	21	44	43
Resilient sectors	20	50	53	30	31
Average of losses in sectors with losses	−12%	−14%	−9%	−5%	−5%
Average of 74 sectors	−8%	0%	3%	0%	−1%

summarizes the number of sectors that reported losses from 2020 to 2024. Typically, 20 sectors face crises, as shown in the table, but during major crises, this number doubles. Between 20 and 30 economic sectors showed great resilience, with no accumulated company exits over the years. The year of COVID-19, 2020, was the year that showed an overall drop of −8% and was followed by 2024, the year of the flood tragedy, where the accumulated loss in activity was 1%. The state’s GDP performance in 2020 aligns with these results, as it dropped 7.2%—the largest decline in two decades.

The analysis of the estimated result versus the actual result, by sector, gives us the value of the estimated annual losses.

Table 9. Performance of selected sectors.

Sectors	2020	2021	2022	2023	2024
Travel agencies, tour operators, and booking services	−27%	4%	2%	18%	−8%
Agriculture, livestock, and related services	−10%	10%	2%	−3%	−3%
Food	−7%	2%	3%	2%	−2%
Accommodation	−9%	4%	3%	5%	−7%
Artistic, creative, and entertainment activities	−63%	−28%	0%	22%	1%
Trade and repair of motor vehicles and motorcycles	−11%	4%	3%	−1%	1%
Wholesale trade	−1%	5%	0%	−2%	0%
Retail trade	−3%	5%	1%	−4%	1%
Manufacture of clothing and accessories	−6%	6%	3%	0%	−2%
Construction of buildings	−1%	−27%	−2%	−3%	14%
Manufacture of beverages	−1%	−4%	3%	2%	2%
Manufacture of machinery and equipment	−16%	1%	13%	0%	−12%
Manufacture of machinery and materials electrical	2%	1%	0%	−2%	1%
Metallurgy	−1%	2%	0%	−1%	−1%
Household services	−20%	8%	10%	18%	−8%

shows the great impact of COVID-19 on the sectors of travel, artistic activities, machinery and equipment industry, and even domestic services. Lockdown policies directly affected many of these sectors by restricting mobility and in-person interactions, but other factors also contributed significantly to the disruption. These include global supply chain interruptions, shifts in consumer behavior, innovation, and the acceleration of digitalization. Together, these elements compounded the economic shock scenario.

In 2024, the year of the flood, the sectors that suffered significant losses were accommodation, travel, machinery and equipment manufacturing, and domestic services. The situation is very similar to that of COVID-19, but of lesser magnitude.

Figure 6 represents the estimated (red line) and actual (blue line) performance of the main economic sectors in Rio Grande do Sul, highlighting the best-performing prediction model and the point of greatest discrepancy (marked by a yellow dot). In eight of the sixteen sectors, this greatest discrepancy occurred during the first wave of COVID-19—with seven indicating losses and one indicating gains. Two additional instances of greatest distance were observed during the flood period.

Specific resilient sectors that outperformed estimates included information services, management consulting, scientific R&D, office support services, and integrated health and social assistance. Vehicle manufacturing also performed well, along with the petroleum products and biofuels sector, and support activities for mineral extraction.

The methods used in this study are complementary, drawing on diverse perspectives to evaluate sectoral performance. Structural breaks are relatively rare, as they capture major disruptions. In this case, they were most evident in manufacturing and transportation—sectors particularly relevant during the COVID-19 pandemic. Several studies highlight how the pandemic disrupted global supply chains. By contrast, few breaks were observed during droughts. Floods, however, triggered structural breaks in only eight sectors, such as information services, computer and electronics manufacturing, optical products, and office and administrative support services provided primarily to businesses.

During the 2024 floods in Rio Grande do Sul, a state of emergency was declared in 323 of the state’s 497 municipalities, while a state of public calamity was declared in another 95. Public calamity represents a more severe status. The disaster directly affected 876,565 people. The most severely impacted regions included the Metropolitan Area of Porto Alegre, the Taquari Valley, and communities surrounding the Patos Lagoon [3].

Several recovery approaches have been identified, reinforcing the findings of the second analysis. One study links the performance of small- and medium-sized enterprises to innovation, showing that the practices adopted to address the repercussions of COVID-19 had a positive impact on business performance and survival prospects [41]. Another study found that tax relief was the most important intervention supporting firms during the pandemic [42].

Other similar studies apply the time to recovery framework as a variation of survival analysis [2,43,44]. The use of survival analysis (SA) is justified both by the way the response variable is constructed and by the fact that the analysis aims to measure duration.

Droughts are frequent events with established mitigation mechanisms, ranging from price adjustments to storage management or buffer stocks, such as financial reserves. Government support mechanisms have proven effective, and because they are frequently activated, their implementation has become increasingly rapid. Instruments that defer tax payments, expand credit, and promote employment retention—particularly among small businesses—are recognized for their strong performance [18].

Research also indicates that most environmental systems exhibit strong resilience to drought events, with impacts generally being temporary [45]. Another side effect of disasters is the increase in inventories and shifts in risk perception [27].

The relatively weak performance of Rio Grande do Sul compared with other states, illustrated in Figure 3 (average of 1.5%), suggests that the long-term effects of recurring natural disasters [1,2] are eroding the state’s overall performance. Moreover, eventual compensation through human capital development may not be sufficient to offset these losses [25].

5. Conclusions

The objective of this study was to assess the extent of the impact that recent natural disasters and pandemics have had on the economic sectors of Rio Grande do Sul.

Three analytical approaches were employed—structural breaks, recovery time, and loss estimation—each contributing distinct and complementary insights to the analysis. The Bai and Perron method effectively identifies structural changes in time series, revealing when significant shifts occur in the data’s behavior. The Kaplan–Meier estimator, widely used in survival analysis, provides a robust framework for assessing recovery time by estimating the duration and distribution of recovery periods following disruptive events. Finally, predictive models evaluated using the Root Mean Square Error (RMSE) offer precise quantification of losses, ensuring the selection of the best-fitting model for accurate impact estimation. Together, these methods form a comprehensive analytical toolkit that captures the timing of disruptions, the resilience of recovery, and the magnitude of losses, enabling a nuanced and thorough understanding of the phenomenon under study.

A total of 82 negative structural breaks were identified, 40 of which were not directly linked to COVID-19, droughts, or flooding events. Notably, 95% of economic activities showed no structural breaks. In other words, large-scale impacts—as indicated by structural breaks—have had only a limited effect on the economy of Rio Grande do Sul. Regarding recovery time for companies that experienced significant losses (greater than 15% of their activity), a similar recovery pattern was observed between the effects of COVID-19 and droughts. By the end of one year, 14% and 15% of affected companies, respectively, remained unrecovered. Overall, two-thirds of companies in the state did not enter a crisis. However, in the case of floods, the proportion of companies affected was higher, reaching 45%.

These findings align with recent studies [2,18]. Nonetheless, the duration that companies remained in crisis varied: approximately six months for climate and hydrological events, and nearly twice as long in the case of COVID-19. The recovery time by sector—measured as the average duration required by companies within each sector to return to pre-crisis levels—indicates that the food sector was the most affected by COVID-19, taking 14 months to recover. During the 2023 droughts, commerce and wholesale trade were the most impacted among the main sectors of the Rio Grande do Sul economy. In the 2024 floods, the food sector once again showed the weakest performance. The sectors most resilient to disasters—capable of maintaining strong performance—include information services, management consulting, research and development, administrative support services, and integrated health and social assistance.

The economy of Rio Grande do Sul has shown notable flexibility. For instance, a water crisis can boost tax revenue by shifting production to thermal energy, which raises prices. In many crises, higher prices help balance lower sales. Companies often adopt costlier, less efficient alternatives that persist due to market demand, keeping invoice levels stable. These findings reinforce previously published studies and contribute to the validation of the existing literature. From a public policy perspective, the results support better targeting of aid during crises and enable improved planning to enhance resilience and sustainability. Sound fiscal policy plays a crucial role in maintaining economic strength, helping to prevent catastrophic projections that might otherwise hinder recovery. Generally, with few exceptions, although some months of hardship may occur, economic stability is typically restored within six months to a year, without major aggregate losses in annual economic flows. One limitation of this study is the limited number of variables considered. We do not explore jobs, average salary, or exports. They could enhance the impact analysis, but they are very complex to add by sector.

Nevertheless, the series spans 100 months and includes more than 15 million observations. Future research should explore the long-term effects of extreme events on factor productivity and examine intersectoral impacts. Despite its limitations, the robust dataset and the development of multiple complementary methods lend strong credibility to this study’s findings.