Slides General Insurance Mathematics
Transcription
Slides General Insurance Mathematics
Dumaria R. Tampubolon, Ph.D Statistics Research Division Faculty of Mathematics and Natural Sciences Institut Teknologi Bandung, Bandung, Indonesia SEAMS School – Universitas Sanata Dharma – Yogyakarta August 2016 Outline What is “Actuarial Science”? What is “General Insurance” and what is its nature? What are the major problems in General Insurance? What is a “runoff triangle”? What is “outstanding claims liability”? Using “Leverage” as a tool to measure the sensitivity of an estimate of outstanding claims liability due to small changes in the incremental payments (paid claims). Determining a probability model for the moment magnitudes of earthquake mainshocks where the source of earthquake is in the region of Megathrust Mid 2 Sumatera Determining the average recurrence interval of certain moment magnitudes. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 2 Actuarial Science A branch of knowledge which applies mathematics, probability and statistics, economics, and finance in assessing risks of financial losses due to measurable non-prevented events. Such events are those which will occur at a certain time in the future with probability greater than zero but the time of occurrence of the event and the corresponding amount of the financial losses cannot be determined with certainty at the time of valuation. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 3 Definition Insurance: A vehicle to transfer pure economic (or financial) risks or liability to compensate for loss or damage arising from specified contingencies such as natural disaster, fire, theft, negligence, injury, death, etc General Insurance: Relates to the insurance of property and liability; may also relate to the insurance of the person which is not covered by life insurance. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 4 Nature of General Insurance Severity (amount of claims) and Frequency (number of claims )can not be determined with certainty at the time of valuation Time of disastrous event which lead to financial compensation can not be determined with certainty at the time of valuation Claims are not usually paid as soon as they occur. Hence, there is a delay: between occurrence of event and reporting of a claim; and between the time of reporting and settlement of the claim. A closed claim might be reopened and additional money need to be paid →Short-tailed and Long-tailed business 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 5 Major Problems in General Insurance How to determine the premium? Risk (Adjusted) Premium = Pure Premium + Loadings Pure Premium is the Expected of Claims, that is the mean of the distribution of claims. How to determine the loading factor? How to determine the outstanding claims reserve? OR How to determine the outstanding claims liability? 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 6 Outstanding Claims Liability The present values of expected future payments which include Incurred But Not Reported (IBNR) claims, the evaluation of future payments on claims already notified and the management expenses of future claims payments of claims incurred as at the balance date. (Hart, D. G., Buchanan, R.A., and Howe, B. A. (1996). The Actuarial Practice of General Insurance, page 26) The provision for the outstanding claims is usually the largest component of a general insurance company’s liabilities; hence, changes in the outstanding claims liability have a direct, and possibly large, impact on the company’s profits and tax liabilities. (General) Insurance companies need to reserve enough of their premium income to cover future claim payments from past and current policies 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 7 Sensitivity of the Estimate Gain insights on the forecasting methodology used: → very or moderately or not sensitive? Gain insights on the data: → absolute and relative importance Gain insights on the uncertainty of the estimate of the outstanding claims liability 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 8 Measurement of Sensitivity Measurement of the sensitivity of the estimate of the outstanding claims liability to small perturbation in an incremental payment: 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 9 EXAMPLE Estimate of the outstanding claims liability of AFG Data using Hertig’s model The data used as an example is the Automatic Facultative General (AFG) Liability, excluding Asbestos and Environmental, from the Historical Loss Development study, which was also considered by Mack (1994b). Following Mack, the runoff triangle of the incurred payments of the AFG data is represented in thousands ($’000). 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 10 EXAMPLE Estimate of the outstanding claims liability of AFG Data using Hertig’s model The estimate of the outstanding claims liability for the AFG data, using Hertig’s Model (Hertig, 1985) is $86.889 million or approximately $87million. What happen if there is an additional $1,000 in the incremental payment on a cell of the runoff triangle? Or, what happen if there is a delay of payment (of a small amount) of $1,000? 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 11 Hertig’s Model Leverage (1 unit increase) 0 1 2 3 4 5 6 7 8 9 0 -1.292 -161.585 -1.352 -0.659 -7.935 -3.322 -13.908 -3.344 2.265 22.815 8-9 August 2016 1 -1.311 1.03 -0.629 -0.469 0.318 0.037 0.367 1.177 2.309 2 -0.513 -1.596 -0.034 0.025 0.254 0.671 1.51 1.664 3 -0.11 0.762 0.257 0.47 0.626 0.842 1.405 4 0.48 0.877 0.643 0.626 0.804 1.454 5 1.201 1.323 1.142 0.996 0.996 6 2.116 2.073 1.677 1.528 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 7 3.237 3.707 2.678 8 5.489 6.455 9 12.161 12 Hertig’s Model Leverage (1 unit increase) 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 13 Sensitivity of the Estimate Given the Hertig’s model, the leverage is calculated as follows. Let us say that there is a $1000 increase in cell (1,0) of the runoff triangle of the incremental payments (an increase of $1000 is small enough since increases of $500 and $1 in the incremental payments also result in the same leverage values). 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 14 Sensitivity of the Estimate The resulting leverage is -161.585. This means that there is a decrease in the estimate of the outstanding claims liability of almost 162 times the increase in the cell. In other words, had the claims paid in accident year 1 and development year 0 been $107,000 instead of $106,000, then the resulting Hertig’s model estimate of the outstanding claims liability will be approximately $162,000 lower than the original estimate of $86,889,000. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 15 Sensitivity of the Estimate In another example, for accident year 0, let us say that there is an increase of $1000 in the paid claims at the final development year (at the tail). Then the change in estimated total outstanding is approximately 12 times as much. This means that, had the $1000 claims been paid later, the resulting Hertig’s model estimate of the outstanding claims liability will be approximately $12,000 higher than the original estimate. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 16 Hertig’s Model Leverage What happens if claim payments are delayed? For a particular accident year: Pay early → a “decrease” in outstanding claims liability estimate Pay later → an “increase” in outstanding claims liability estimate What happens where there are very few observations available to do forecasting? Large leverage in the last accident year and at the tails There is an extremely large leverage in cell (1,0). It turned out that the incremental payment in that cell is “unusual” compare to the other incremental payments in development year 0. The leverage values of the Hertig’s model indicate “unusual observations” in the incremental payment data. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 17 Earthquake Insurance How to determine the premium to cover loss due to earthquake hazard? How to determine the outstanding claims liability? There is an expression used in seismology and geophysics ; for example: “10% PE in 50 years given the return period of 475 years “. What does that mean? 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 18 Determining Premium using Catastrophe Model for Earthquake Let say someone asks an insurance company to provide protections against earthquakes; that is a coverage on damages to a property or on business’ interruption caused by an earthquake. What is the premium need to be charged by the insurance company? To determine the premium, one need to estimate the financial loss caused by the earthquake; or one need to determine the “probable maximum loss”; or one need to determine the probability of the loss to exceed a certain amount. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 19 Peta Tim Revisi Gempa 2010 2010 Indonesia Hazard Map (Open a Different File) 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 20 Significant earthquakes in Indonesia 2004 – 2011 (Source: USGS website) 26 December 2004 Sumatera-Andaman Islands (Aceh): Mw 9.1 (tsunami); 227,898 fatalities 28 March 2005 Northern Sumatera (Nias region): Mw 8.6; 1,313 fatalities 27 May 2006 Java (Yogyakarta): Mw 6.3; 5,749 fatalities 12 September 2007 Southern Sumatera (Bengkulu): Mw 8.5; 25 fatalities 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 21 Significant earthquakes in Indonesia 2004 – 2011 (Source: USGS website) 16 November 2008 North Sulawesi (Gorontalo): Mw 7.4; 6 fatalities 30 September 2009 Southern Sumatera (Padang): Mw 7.6; 1,117 fatalities 25 October 2010 Kepulauan Mentawai : Mw 7.7 (tsunami); 670 fatalities 4 April 2011 South Java: Mw 6.7 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 22 Catastrophe Model -Earthquake (Moment) Magnitude -Distance between the Site and the Source of Earthquake -Soil Condition HAZARD VULNERABILITY LOSS INVENTORY 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 23 Case Study Megathrust Mid 2 Sumatera 97.298°E – 101.947°E and -5.418°S – 0.128°N West Sumatera Province, Indonesia (Irsyam et al., 2010) 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 24 Modeling Hazard: Moment Magnitude Gutenberg-Richter Law Gutenberg-Richter (1941; 1944) describe the relationship between the frequency and earthquake magnitude through the equation log10 𝑁 𝑚 = 𝑎 − 𝑏𝑚 where N(m) is the number of earthquakes with magnitudes greater than or equal to m; a and b are parameters which indicate the characteristics of seismic activities at a particular site. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 25 Modeling Hazard: Moment Magnitude In practice, it is of interest to examine earthquake magnitudes which are greater than or equal to a particular value mt . Hence, the Gutenberg-Richter equation becomes log10 𝑁 𝑚 = 𝑎𝑡 − 𝑏 𝑚 − 𝑚𝑡 or 𝑁 𝑚 = 10𝑎 𝑡 −𝑏 𝑚 −𝑚 𝑡 where m is greater than or equal to mt ; and at is the logarithm of the number of earthquakes with magnitude greater than or equal to mt . 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 26 Modeling Hazard: Moment Magnitude In this research, we will use the moment magnitude scale instead of the Richter scale to measure the earthquake magnitude. Let Mw be a random variable which denote the moment magnitude of an earthquake. Let Z be a random variable which denote the scalar seismic moment in Newton-meter or Nm. The relationship between earthquake moment magnitude and the scalar seismic moment is 2 𝑀𝑤 = 3 log 𝑍 − 6 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 27 Modeling Hazard: Moment Magnitude Hence, the number of earthquakes with Mw greater than or equal to M is given by the equation 𝑁 𝑀 = 10𝑎 𝑡 −𝑏 or 𝑁 𝑀 = 10 𝑎 𝑡 −𝑏 𝑀−𝑀𝑡 2 2 log 𝑍− log 𝑍𝑡 3 3 Since at = log10 N(Mt) then it can be shown that the equation above is equivalent to: 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 28 Modeling Hazard: Moment Magnitude 𝑁 𝑀 = 𝑁 𝑀𝑡 where 𝑍𝑡 𝑍 𝛽 2 𝛽= 𝑏 3 Hence, 𝑍𝑡 Pr 𝑀𝑤 > 𝑀 𝑀𝑤 ≥ 𝑀𝑡 = 𝑍 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 𝛽 29 Modeling Hazard: Moment Magnitude So, given Mw greater than or equal to Mt , the equation above is the survival function of a Pareto distribution with parameters β and Zt . That is, given the moment magnitude is greater than or equal to a threshold Mt , the seismic moment Z follows a Pareto distribution with parameters β and Zt . This result leads to a hypothesis that the moment magnitude of earthquake mainshocks might follow a Generalized Pareto distribution. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 30 Modeling Hazard: Moment Magnitude The data used in this research is the earthquake data from the National Earthquake Information Centre – United States Geological Survey (USGS, 2012) earthquake catalog, from January 1973 to December 2011. The selected earthquake data are those of which centre are in the area of Megathrust Mid 2 Sumatera. We use the report by “Tim Revisi Peta Gempa Indonesia tahun 2010” (Irsyam et al, 2010) in defining the area of Megathrust Mid 2 Sumatera. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 31 Modeling Hazard: Moment Magnitude The earthquake mainshocks are separated from the earthquake foreshocks and aftershocks. The process is called seismicity declustering. The Gardner – Knopoff algorithm (1974) and the program written by Stiphout et al (2012) are used to decluster the earthquakes data. After declustering, 6.82% or 137 earthquakes are categorized as earthquake mainshocks. The descriptive statistics of the earthquake mainschoks data are as follows: 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 32 Modeling Hazard: Moment Magnitude N Mean Variance Standard Deviation Skewness Kurtosis Lower Quartile Median Upper Quartile Range 8-9 August 2016 137 5.6814 0.2006 0.4479 3.0609 14.4983 5.4146 5.5727 5.8155 3.4066 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 33 Modeling Hazard: Moment Magnitude We fit a Generalized Pareto distribution to the moment magnitudes of the earthquake mainshocks data. The distribution function of a Generalized Pareto distribution with parameters ξ and θ, and threshold u is: 𝑥−𝑢 1− 1+𝜉 𝐹𝑋 𝑥 = 𝜃 1−𝑒 8-9 August 2016 − 𝑥−𝑢 𝜃 1 − 𝜉 , , if 𝜉 ≠ 0 , 𝜃 > 0 if 𝜉 = 0 , 𝜃 > 0 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 34 Modeling Hazard: Moment Magnitude To estimate the parameters, the Maximum Likelihood Estimation method is applied. At 5% significance level, the Cramér-von Mises test statistics showed that, given a threshold of moment magnitude Mt = 5.4, the moment magnitudes of the earthquake mainshocks (with the source of earthquakes in the area of Megathrust Mid 2 Sumatera) follows a Generalized Pareto distribution with parameters ξ = 0.14447 and θ = 0.30891. With the parameters obtained, the distribution of the moment magnitudes has mean 5.76107 and standard deviation 0.42820. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 35 Modeling Hazard: Average Recurrence Interval 𝑀𝑤 be a random variable which denote the moment magnitudes of earthquakes (mainshocks) 𝑉 be a random variable which denote the number of years needed until an earthquake with moment magnitude at least a certain value, 𝑀𝑤 ≥ 𝑀𝑡 , occurs for the first time. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 36 Modeling Hazard: Average Recurrence Interval V follows a Geometric distribution with parameter 𝑝 = Pr 𝑀𝑤 ≥ 𝑀𝑡 The expected number of years needed until an earthquake with Mw at least Mt occurs for the first time is 1 𝐸𝑉 = Pr 𝑀𝑤 ≥ 𝑀𝑡 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 37 Modeling Hazard: Average Recurrence Interval The “average recurrence interval“ (some literature used the term “return period”) is defined as the expected number of years until an earthquake with Mw at least M, given a moment magnitude threshold Mt , occurs for the first time in a region. Average Recurrence Interval = 𝜏 = Pr 𝑀 1 𝑤 >𝑀 𝑀𝑤 >𝑀𝑡 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 38 Modeling Hazard: Average Recurrence Interval 1 The value is called the “average recurrence rate”. 𝜏 Let N be the random variable which denote the number of earthquakes with Mw at least M, given a moment magnitude threshold Mt , occurring in t years in a region. It is assumed that the earthquake (mainshock) is independent of time and independent of past earthquakes (mainshocks). 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 39 Modeling Hazard: Average Recurrence Interval The random variable N may be modeled by a Poisson distribution, that is Pr 𝑁 = 𝑛 = 𝑡 − 𝑒 𝜏 𝑛! 𝑡 𝜏 𝑛 for n = 0,1,2,… 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 40 Modeling Hazard: Average Recurrence Interval The probability of at least one earthquake with Mw at least M, given a moment magnitude threshold Mt , occurring in t years in a region is: Pr 𝑁 ≥ 1 = 1 − 𝑒𝑥𝑝 𝑡 𝜏 The above equation can be used to calculate the seismic risk expressed as: “x% PE in t years” (x% Probability of Exceedance in t years) for a given recurrence interval of earthquakes with a certain moment magnitude or greater. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 41 Modeling Hazard: Average Recurrence Interval Example: Let t = 50 years and let the probability of exceedance 0.1 Then 0.1 = 1 − 𝑒𝑥𝑝 − 50 𝜏 or the average recurrence interval is approximately 475 years 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 42 Modeling Hazard: Averange Recurrence Interval For a threshold Mt = 5.4, let Mw follows a Generalized Pareto distribution with parameters 𝜉 = 0.14447 and Let 𝜃 = 0.30891 Pr 𝑀𝑤 > 𝑀 = 𝑆𝑀𝑤 1 𝑀 = 475 Then M is approximately 8.475 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 43 Modeling Hazard: Averange Recurrence Interval This means that “the probability of at least one earthquake with moment magnitude at least 8.475 occurring in 50 years in the region, given the average recurrence interval of 475 years, is 10%”. Using the expression usually used by seismologists: “10% PE in 50 years given the average recurrence interval of 475 years with moment magnitude at least 8.5 “. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 44 Modeling Hazard: Averange Recurrence Interval Another example: The probability of at least one earthquake with moment magnitude at least 7.935 occurring in 50 years in the region, given the average recurrence interval of 224 years, is 20%. Using the expression usually used by seismologists: “20% PE in 50 years given the average recurrence interval of 224 years with moment magnitude at least 7.9 “ 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 45 Still Need to be Done!!! Modeling the distribution of Peak Ground Acceleration (PGA) utilizing the probability distribution of moment magnitudes (Hazard Module) Determining the Modified Mercalli Intensity (MMI) Determining the Damage Curve (Vulnerability Module) Determining the “Probable Maximum Loss”; or determining the Distribution of Loss (Loss Module). Determining the Premium Determining the Outstanding Claims Liability 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 46 References [1] [2} [3] [4] Andaria, R. (2013). Penentuan Average Recurrence Interval, Peak Ground Acceleration dan Modified Mercalli Intensity: Sumber Gempa Wilayah Megathrust Mid 2 Sumatera, Tesis Program Studi Magister Matematika (Supervisor: Tampubolon, D. R.), FMIPA, Institut Teknologi Bandung. Chen, W. and Scawthorn, C. (2003). Earthquake Engineering Handbook. CRC Press. Choulakian, V. and Stephens, M. (2001). “Goodness-of-Fit Tests for the Generalized Pareto Distribution”. Technometrics, American Statistical Association and American Society for Quality Control, 43, 4, 478-484 Hertig, J. (1985), "A Statistical Approach to IBNR-Reserves in Marine Reinsurance”, ASTIN Bulletin, 15, 2, 171-183. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 47 Reference [5] Irsyam, M., Sengara, I. W., Aldiamar, F., Widiyantoro, S., Triyoso, W., Natawidjaja, D. H., Kertapati, E., Meilano, I., Suhardjono, Asrurifak, M., and Ridwan, M. (2010). Ringkasan Hasil Studi Tim Revisi Peta Gempa Indonesia 2010, Technical Report, Departemen Pekerjaan Umum, Indonesia. [6] Kagan, Y. (2002). “Seisimic Moment Distribution Revisited: I. Statistical Results”, Geophysical Journal International, 148, 520-541 [7] Klugman, S., Panjer, H., Willmot, G. (2004). Loss Models: From Data to Decisions, 2nd edition, New York: Wiley. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 48 Reference [8] Mack, T. (1994b), "Measuring the Variability of Chain Ladder Reserve Estimates”, Casualty Actuarial Society Forum, Spring, 101-182. [9] Pisarenko, V., Sornette, A., Sorenette, D., and Rodkin, M. (2008). “Characterization of the Tail of the Distributions of Earthquake Magnitudes by Combining the GEV and GPD Descriptions of Extreme Value Theory”, http://arxiv.org/ftp/arxiv/papers/0805/0805.1635.pdf [10] Pradana, A. A. (2013). Pemodelan Magnitudo Gempa Bumi Menggunakan Distribusi Peluang Generalized Pareto: Studi Kasus Megathrust Mid 2 Sumatera dan Megathrust Jawa, Laporan Tugas Akhir Program Studi Sarjana Matematika (Supervisor: Tampubolon, D. R.), FMIPA, Institut Teknologi Bandung. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 49 Reference [11] Putra, R. R., Kiyono, J., Ono, Y., dan Parajuli, H. R. (2012). “Seismic Hazard Analysis for Indonesia”, Journal of Natural Disaster Science, 33, 2, 59-70. [12] Stiphout, T., Zhuang, J., and Marsan, D. (2012). “Seismicity Declustering”, Community Online Resource for Statistical Seismicity Analysis. [13] Tampubolon, D. R. (2008). Uncertainties in the Estimation of Outstanding Claims Liability in General Insurance, PhD Thesis, Macquarie University, Australia 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 50 Reference [14] Tse, Y. K. (2009). Non-life Actuarial Models: Theory, Methods and Evaluation, New York: Cambridge University Press. [15] United States Geological Survey (USGS) website. Earthquake search. http://earthquake.usgs.gov/earthquakes/eqarchives [16] Wang, Z. (2007). “Seismic hazard and risk assessment in the intraplate environment: The New Madrid seismic zone of the central United States”, The Geological Society of America. Special Paper 425, 363-373. 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 51 Thank You 8-9 August 2016 D. R. Tampubolon: SEAMS School at Universitas Sanata Dharma 52