ECON 107 Econometrics and Catastrophes in South East Asia Sijie , Derek, Shu Jun Hao, Alvinder , Teo Heng Kai Instead of combining obvious variables linking to costs of floods, we sieve out significant yet under - looked factors to drive results Research Methodology How did we do it? Finalising Objectives: Analysing financial impact on Indonesia due to flood ”Data Mining” – Finding abnormal explanatory variables that contributes to costs of floods Single linear regressions Forming the big picture – How much do these variables explain the variations Multiple linear regression Interpreting the regression results – Does it make sense? Relevant tests Objectives Single Linear Regressions Multi Linear Regressions Overall Results Natural Disaster Selection Introduction on Catastrophes in SEA Which Country, What Disaster? Looking ahead, floods risk will continue to increase and face greater economic impact, especially in SEA Objectives Single Linear Regressions Multi Linear Regressions Overall Results - Within Southeast Asia, frequency of hydrometeorological disasters (floods and storms) are increasing more rapidly than geophysical disasters (earthquakes, tsunamis and volcanic eruptions) - Despite higher economic damages registered by both earthquake and drought, flood damages are higher due to its high frequency of occurrence - With rising sea level, hydrometeorological disasters continue to be a key concern – in Asia, 70% of 200M of humans will be affected a majority resides in Southeast Asia 0 5 10 15 20 25 30 35 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 Biolog ical Geophysical Hy drological Meteorolog ical Climatolog ical Frequency of Natural Disasters in SEA (5MA) Source: EM - DAT 291,333 601,923 94,942 200,059 10,372 155,226 200,388 0 100,000 200,000 300,000 400,000 500,000 600,000 700,000 Drought Earthquake Extreme temperature Flood Volcano Storm Wildfire Source: PreventionWeb Economic Damage by Event (US $’000s) Introduction on Catastrophes in SEA Which Country, What Disaster? By analysing causes of economic damages due to flooding in Indonesia, we hope to provide insights on factors resulting in highest damage - Apart from being located on the Pacific Ring of Fire, Indonesia is extremely vulnerable to flooding due to its low - lying coastal areas - In 2019, Indonesia announced to move its capital from Jakarta to a more central location – acknowledging Jakarta will be submerged entirely by 2050 - Evaluation of future flood risk revealed that flood costs would increase by 400% by 2050, mainly due to: 1. Rapid developments in Indonesia 2. Ongoing climate crisis Objectives Single Linear Regressions Multi Linear Regressions Overall Results Source: Statista Region in SEA - Based on F test on goodness of fit and T - test, reject null hypothesis at 5% significance and the model has explanatory power - This meant that when Ageing Population increases by 1000 , the costs of disasters increases by US$51,819.53 - The R 2 is relatively low at 11.7% - Based on Durbin - Watson stat of 1.935 , we can conclude that there is no serial correlation at 1% significance level [d L : 1.288 & d U : 1.376] - This is evident in the residual plot as well Simple Linear Regression Explanatory Variables: Ageing Population F & t Tests are significant at 5% significance & intuitively the larger the ageing population, the more costly the disasters wou ld be Objectives Single Linear Regressions Multi Linear Regressions Overall Results Intuitive Explanation of Variable Tests Statistics of Regression EViews Results Residual Plot 0 500 1000 1500 2000 2500 3000 Residuals - Intuitively, the older adults are among the most vulnerable groups to natural disasters due to decreased sensory awareness, physical impairment, chronic medical conditions 1 - Therefore, they are more susceptible to long - term care for physical or psychological recovery or even death - This would likely cause the increase in costs of disasters due to possibly higher healthcare costs or death rates Source: National Center for Biotechnology Information Source: Passport Simple Linear Regression Explanatory Variables: Total Affected F & t Tests are significant at 10% significance & intuitively the more people affected, the more costly the disasters would b e Objectives Single Linear Regressions Multi Linear Regressions Overall Results EViews Results Residual Plot 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 Residuals - Intuitively, the greater the number of people affected, the greater the damage of floods (in terms of costs ) - However, since R 2 is still relatively low, we can conclude that the number of people affected does not fully explain the costs of disasters and there may be other variables affecting the costs - Based on F test on goodness of fit and T - test, reject null hypothesis at 10% significance and the model has explanatory power - This meant that when Total Affected increases by 1000 , the costs of disasters increases by US$365,168.00 - The R 2 is relatively low at 7.2% - Based on Durbin - Watson stat of 1.697 , we can conclude that there is no serial correlation at 1% significance level [d L : 1.288 & d U : 1.376] - This is evident in the residual plot as well Source: Passport Intuitive Explanation of Variable Tests Statistics of Regression Simple Linear Regression Explanatory Variables: Urbanization F & t Tests are significant at 2.5% significance & intuitively the greater the degree of urbanization, the more costly the di sas ters would be Objectives Single Linear Regressions Multi Linear Regressions Overall Results EViews Results Residual Plot 0 20000 40000 60000 80000 100000 120000 140000 160000 180000 Residuals - As there is no direct measure of urbanization, we have used urban population as a proxy for the degree of urbanization - Based on F test on goodness of fit and T - test, reject null hypothesis at 2.5% significance and the model has explanatory power - This meant that when Urban Population increases by 1000 , the costs of disasters increases by US$5219.45 - The R 2 is relatively low at 12.7% - Based on Durbin - Watson stat of 1.953 , we can conclude that there is no serial correlation at 1% significance level [d L : 1.288 & d U : 1.376] - This is evident in the residual plot as well - Studies 1 have shown that flooding disasters are closely linked to rapid and unchecked urbanization which forces low - income families to settle on the slopes of steep hillsides or ravines, or along the banks of flood - prone rivers - The massive numbers of urban poor have fewer options for availability of safe and desirable places to build their houses - Thus, urbanization is likely to lead to increase in costs due to the urban poor having no choice but to stay along flood - prone rivers Source: Department of Humanitarian Affairs/United Nations Disaster Relief Office Source: Passport Intuitive Explanation of Variable Tests Statistics of Regression Multi Linear Regression Interpreting the results Objectives Single Linear Regressions Multi Linear Regressions Overall Results Coefficient of Ageing Population - Based on gut feeling, an older population should result in a higher costs as floods occur compared to a younger one – idea that young is able to “escape” far better than elders - Yet empirically, coefficient of aging population is calculated to be - 225.8476, per thousand increase in elders decreases total damages by $225.8476USD - This contrasts with the earlier single linear regression where the coefficient produced was positive. - It is interesting to observe this phenomenon as we see the effect of an ageing population in a multi regression framework. - Its effect in this multi regression changes sign, showing how in a joint explanatory manner, that an increasingly aged population reduces the total damage, holding other factors constant. EViews Results 𝑇𝑜𝑡𝑎𝑙 𝑑𝑎𝑚𝑎𝑔𝑒𝑠 = 𝛽 ! + 𝛽 " ∗ 𝑇𝑜𝑡𝑎𝑙 𝐴𝑓𝑓𝑒𝑐𝑡𝑒𝑑 + 𝛽 # ∗ 𝐴𝑔𝑒𝑖𝑛𝑔 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 + 𝛽 $ ∗ 𝑈𝑟𝑏𝑎𝑛 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 - Studies have further supported this negative coefficient contrary to our initial beliefs. - There are massive contributions from the part of the elderly, acting as pillars of support for their family units and communities during times of crisis - Their past experiences and “accumulated knowledge” from prior disasters prove to be a source of learning and development for the younger crowd. - In fact, the Elderly play a vital role in emergency preparedness and rapid response in times of disaster. Source: World Health Organisation Intuitive Explanation of Variable Multi Linear Regression Interpreting the results F Test is significant at 5% significance, however due to multicollinearity problem, t statistics of ‘Ageing Population’ and ‘ Urb an Population’ are insignificant Objectives Single Linear Regressions Multi Linear Regressions Overall Results Tests Statistics of Regression - Based on F test of goodness of fit, reject null hypothesis at 5% significance and the model has explanatory power 𝑭 𝟒 − 𝟏 , 𝟒𝟒 − 𝟒 = 𝟑 𝟐𝟑𝟏 > 𝑭 𝒄𝒓𝒊𝒕 𝟓% 𝟑 , 𝟒𝟎 = 𝟐 𝟖𝟒 - However, the model suffers from severe multicollinearity given the high correlation of 99.4% between ’Ageing Population’ and ‘Urban Population’ - Individual t test statistics are now insignificant (beyond 10% alpha) since standard errors have increased – loss of precision EViews Results correlation Aging Population Urban Population Aging Population 1 0.994 Urban Population 0.994 1 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 1977 1984 1991 1998 2005 2012 2019 Millions Millions Urband_Pop Ageing_Pop Ramsey Reset Test of Misspecification to check for non - linearity t - statistic of 𝐹𝐼𝑇𝑇𝐸 𝐷 𝟐 is insignificant, hence, there is no non - linearity and our model is likely to be correctly specified Objectives Single Linear Regressions Multi Linear Regressions Overall Results Multi Linear Regression EViews Results: 𝑑𝑓 = 44 − 4 = 40 , 𝑡 %&'( , * % = 1 96 𝑡 ,(-( < 𝑡 %&'( , * % 𝑓𝑜𝑟 𝐹𝐼𝑇𝑇𝐸 𝐷 " - Ramsey’s Reset Test yields an insignificant t - statistic of 0 9474 for the 𝐹𝐼𝑇𝑇𝐸 𝐷 ( variable at 5 % significance level - This indicates the absence of non - linearity in our regression model - Our model is likely to be correctly specified despite its weak explanatory power - The analysis is backed by the F test of goodness of fit as shown previously Results Analysis Possible Direct Measures to alleviate Multicollinearity Effectiveness of measures in our model All direct measures cannot be used to alleviate the multicollinearity problem in our model Objectives Single Linear Regressions Multi Linear Regressions Overall Results Possible Methods Why these methods are inapplicable 1. Increase the number of observations (n) 1. No availability of additional real - life data that further improves the model 2. Reduce the variance of the disturbance term 2. It is hard to reduce the variance of the disturbance terms because there is a high chance of including irrelevant or correlated variables that can potentially increase the standard errors 3. Increase the Mean Square Deviation (MSD) 3. Increasing MSD can only be done in the early design stages 4. Reduce the correlation between explanatory variables, r 2 X2,X3 4. Due to the three explanatory variables used, it is hard to further decrease the correlation without affecting the validity of the model, and it can only be done in the early design stages Possible Indirect Measures to alleviate Multicollinearity Although multicollinearity problem seems to be alleviated, we run into the problem of omitted variable bias and conclude that th is method does not work. Objectives Single Linear Regressions Multi Linear Regressions Overall Results Rationale 1. Dropping Urban Population Variable - High correlation coefficient of 99 4 % & lower coefficient value - Standard error of ‘Ageing Pop’ fell, multicollinearity seems to be alleviated - ‘Total Affected’ has become insignificant at 10 % significance levels for both, but the F stats for both models are significant at least at 5 % significance - There may be omitted variable bias because 𝑅 ( is likely to be overinflated The remaining variable likely to be acting partly as a proxy for the dropped variable, inflating its apparent explanatory power - Having 3 variables is a better choice because dropping a variable may lead to omitted variable bias Figure 1: Dropping Urban Population Results Analysis 𝑇𝑜𝑡𝑎𝑙 𝐷𝑎𝑚𝑎𝑔𝑒𝑠 = 𝛽 ! + 𝛽 " ∗ 𝑇𝑜𝑡𝑎𝑙 𝐴𝑓𝑓𝑒𝑐𝑡𝑒𝑑 + 𝛽 # ∗ 𝐴𝑔𝑒𝑖𝑛𝑔 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 + 𝛽 $ ∗ 𝑈𝑟𝑏𝑎𝑛 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑇𝑜𝑡𝑎𝑙 𝐷𝑎𝑚𝑎𝑔𝑒𝑠 = 𝛽 ! + 𝛽 " ∗ 𝑇𝑜𝑡𝑎𝑙 𝐴𝑓𝑓𝑒𝑐𝑡𝑒𝑑 + 𝛽 # ∗ 𝐴𝑔𝑒𝑖𝑛𝑔 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝐹 <=>? , A% 2 , 44 − 3 ≈ 𝐹 <=>? , A% 2 , 40 = 3 23 < 3 792 ( 𝐹 − 𝑆𝑡𝑎𝑡 ) Figure 2: Multi Regression Model Possible Indirect Measures to alleviate Multicollinearity Objectives Single Linear Regressions Multi Linear Regressions Overall Results EViews Results: - We put a linear restriction on the two factors and combined them into one variable AGEURBAN Rationale 2. Linear Restriction on Ageing Population and Urban Population Results Analysis - AGEURBAN is significant at 5 % significance level - ‘Total Affected’ has become insignificant at 10 % significance levels, but the F stat is significant at 2 5 % significance level F - Test on Linear Restriction 𝑇𝑜𝑡𝑎𝑙 𝑑𝑎𝑚𝑎𝑔𝑒𝑠 = 𝛽 ! + 𝛽 " ∗ 𝑇𝑜𝑡𝑎𝑙 𝐴𝑓𝑓𝑒𝑐𝑡𝑒𝑑 + 𝛽 # ∗ 𝐴𝑔𝑒𝑖𝑛𝑔 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 + 𝛽 $ ∗ 𝑈𝑟𝑏𝑎𝑛 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝐹 1 , 44 − 4 = 9 69 ∗ 10 !" − 9 37 ∗ 10 !" ( 9 37 ∗ 10 !" ) / 40 = 1 366 < 𝐹 %&'( , *% 1 , 40 = 4 08 𝑇𝑜 𝑡𝑒𝑠𝑡 ∶ 𝐻 ) : 𝛽 * = 𝛽 + 𝑣𝑠 𝐻 , : 𝛽 * ≠ 𝛽 + Do not reject 𝐻 ) at 5 % significance level, the linear restriction is valid Linear restriction has only alleviated the multicollinearity problem, but ‘Total Affected’ has become insignificant even thou gh the model has explanatory power. 𝐹 <=>? , " A% 2 , 44 − 3 ≈ 𝐹 <=>? , " A% 2 , 40 = 4 05 < 4 121 ( 𝐹 − 𝑠𝑡𝑎𝑡 ) Conclusion Key Takeaways Multi - regression model suggests prioritizing elders’ knowledge to drive greater impact in reducing flood costs Objectives Single Linear Regressions Multi Linear Regressions Overall Results Validity of Multi - regression Model - F - test on goodness of fit is significant at 5% significance level – implies that the model has explanatory power • Here, other (more commonly relevant) variables are ignored. With the objective focused on abnormal variables contributing to costs of floods, 𝑅 ( will be lower than traditionally accepted - Ageing Population and Urban Population variables have insignificant t statistics at 10% significance level – indicates severe multicollinearity problem • However, coefficients of variables remain unbiased and accurate in explaining the relationship with the costs of flood - Analysing the magnitude each factor poses in contributing to the extent of financial damages due to flood • Intuitively, a positive relationship exists between cost and total affected – reducing total affected will bring down the costs of floods as healthcare cost is a main attribute • Positive relationship between costs and urban population - by relocating Indonesians out of densely populated regions, floods costs will decrease • Interestingly, the negative relationship between proportion of ageing population and flood costs implies that the elders are significant in helping reduce future costs – years of experience and knowledge enable elders to better manage the disaster à Relevant authorities should involve elders in devising disaster prevention and control measures – by tapping on their experiences, Indonesians will smoothen the learning curve of dealing with future floods, reducing total costs • Rafiey , H., Momtaz , Y. A., Alipour , F., Khankeh , H., Ahmadi, S., Sabzi Khoshnami , M., & Haron , S. A. (2016). Are older people more vulnerable to long - term impacts of disasters?. Clinical interventions in aging , 11 , 1791 – 1795. https://doi.org/10.2147/CIA.S122122 • Basnyat , B. (2009, August). IMPACTS OF DEMOGRAPHIC CHANGES ON FORESTS AND FORESTRY IN ASIA AND THE PACIFIC . Home | Food and Agriculture Organization of the United Nations. https://www.fao.org/3/am253e/am253e.pdf • Hutton, D. (2008, October). Older people in emergencies: Considerations for action and policy development . WHO/OMS: Extranet Systems. https://extranet.who.int/agefriendlyworld/wp - content/uploads/2014/06/WHO - Older - Persons - in - Emergencies - Considerations - for - Action - and - Policy - Development - English.pdf • WHO. (2008). OLDER PERSONS IN EMERGENCIES: AN ACTIVE AGEING Perspective . WHO | World Health Organization. https://www.who.int/ageing/publications/EmergenciesEnglish13August.pdf • Passport. (n.d.). Population Aged 65+ https://www - portal - euromonitor - com.libproxy.smu.edu.sg/portal/StatisticsEvolution/index • Passport. (n.d.). Indonesia Total Population https://www - portal - euromonitor - com.libproxy.smu.edu.sg/portal/StatisticsEvolution/index • Passport. (n.d.). Urba population living in Indonesia https://www - portal - euromonitor - com.libproxy.smu.edu.sg/portal/statisticsevolution/index • Disaster statistics - Asia - Countries & regions - PreventionWeb.net. (n.d.). PreventionWeb - Knowledge platform for disaster risk reduction. https://www.preventionweb.net/english/countries/statistics/index_region.php?rid=4 • Total affected . (n.d.). https://public.emdat.be/data • Total cost . (n.d.). https://public.emdat.be/data Sources Objectives Single Linear Regressions Multi Linear Regressions Overall Results