Big Data: A new era of insurance? Dr. Iordanis Chatziprodromou, Mexico 2015

Transcription

1 Big Data: A new era of insurance? Dr. Iordanis Chatziprodromou, Mexico 2015

2 What is Big Data? 2

3 What is Big Data? There are several definitions for Big Data based on different properties regarding the technology and analytics needed to encapsulate the analysis: Data volume of an average size company >427x > Size based: " data that is too big to fit on a single server" Type based: " too unstructured to fit into a row-and-column database" 90% of the world data created the last 2 years Internet based information is doubling every 18 months Pace based: " too continuously flowing to fit into a static warehouse" Uncertainty based: " too uncertain to fit into any traditional analytical framework" Traditional data Numerical only Text Pictures... Sensoric Data Audios Videos 3

4 Switch these single quotes to double quotes, or eliminate Our definition Big Data is data with "high" Volume "high" Velocity "high" Variety "high" Veracity "high" = anything that could not be managed by traditional technological infrastructure and analytical methods 4

5 Three key mind-set changes Traditional analysis Use of Samples N<All Very clean data N<All because we are using samples for the analysis, the results should be extrapolated to the population in order to draw conclusions with a subsequent amplification of the estimation error Causations A B, B happens because of A Big Data analysis Use of whole populations N=All Messier data N=All No extrapolation is needed and thus no subsequent amplification of the error. This property allows us to tolerate higher error that is coming from the use of the messier Big Data Correlations A B, A and B are happening together with no clear causal relation 5

6 Insurance examples 6

7 General insurance examples Contingent Business Interruption (CBI) Getting insights into a combination of geography data, accumulation of hazards and industry parks is key to developing critical scenarios and to identifying interdependencies. Big Data can help to enhance risk management in the context of supply chain activities and to better map risk correlations as well as to improve the management of accumulation control. Fraud Detection The estimation of insurance losses due to fraudulent claims sums up to a staggering $30billion every year. Pattern recognition methods applied on insurance data and relevant external Big Data, will help reduce this amount significantly. Telematics By fitting an electronic device into vehicles which captures and transmits information, a completely new source of data is provided to the insurer enabling usage based insurance as well as improving the loss ratio by improving driving behaviour. Moreover, additional services, such as how to change or adapt the driving style to drive more ecologically can be offered to the policy holder. 7

8 Claims specific examples First Loss Estimate In the case of a catastrophe, a combination of information coming from mobile devices - like vertical displacement sensors or barometers - social media, satellite pictures etc., can be used in order to accurately and timely assess the extend of a certain loss. Consequently, the quality of the claims management and the reserving process will improve significantly Forensics A lot of times small to mid size events are falling off the insurance radar due to the lack of global media coverage. A systematic monitoring of regional media in local languages will enable insurers with a information which, among other benefits, will help them to proactively manage their claims and remove uncertainties around the extent of "grey" zone claims Systemic Risk Change Identification Our world is changing rapidly and we experience abrupt breaks of structural risk patterns due to new technological enhancements. Traditional insurance models will fail to timely capture the new phenomenon due to their need for a claim history in order to adopt to the new risk "normal" with potentially very adverse effects. With the use of Big Data analytics, that are by definition much faster, this risk can be mitigated. 8

9 Key thoughts 9

10 Key Big Data related thoughts 1. We live in a new technological era that demands new types of analytics 2. We are at the disruptive transitional phase that requires changes and investment that can be painful short term but is necessary for the evolution of the business 3. Big Data science is NOW the equivalent of what Actuarial science was in the 90s for insurance Might help to note who "The general understanding is that early adopters of Big Data will probably thrive, and companies that fail to you are quoting here. may struggle to survive. Swiss Re believes that Big Data is an opportunity to further differentiate and to add value to the business." 10

11 Global Motor Risk Map

12 Prelude 12

13 What The Team does Uses different non insurance data Builds different non insurance models Derives insurancerelevant unique insights related to motor risk for accidents 13

14 The Data 14

15 What kind of data and where to find it? Mexico Geo Data extraction 1. Raster of population data (ORNL) Population density, standard deviation 2. Image processing of satellite images of Land cover data (ESA) Area of forest, Rural, Urban, Crop lands, etc. 3. Image processing of satellite images of night light data (NOAA) Night light value (proxy measure of Human Well-Being such as GDP) 4. Data extraction from map data Road type, length and characteristics (HERE) e.g. highways, primary roads, trunk roads, pedestrian roads, etc. Road junctions, one ways, tunnels, bridges and speed limits 5. Image processing of satellite images of precipitation data (ECMWF) Average yearly rainfall and total rainfall 6. Image processing of satellite images of digital elevation model (NASA) Maximum elevation, standard deviation for elevation and mean elevation 15

16 Night light intensity extraction from satellite images an example Source This is an indicator of economic factors such as GDP * * Bleakley and Lin (2012), Henderson et al. (2012)1, Michalopoulos and Papaioannou (2013), Lowe (2014), Storeygard (2012), Pinkovskiy (2013) 16

17 The Model 17

18 Country A Country B Country C Model description Data on the different features (road network, population density, etc.) + real accident data Data to calibrate model Usage of a regression tree to separate different groups of regions model for different types of regions Mexico feature data Group 1 Group n Model Different model approaches tested Poisson regression Generalised linear model Linear models per group of regions prediction of accident frequency per capita 18

19 Results & analysis 19

20 Colombia risk map Predicted accident frequency per capita Mexico City Usage of high-quality road network data improved the model by more than 50% 20

21 density Risk characteristics two different cases San Luis Potosí Tamaulipas Let s have a closer look at two departments and see what our model says: Tamaulipas High risk San Luis Potosí Low risk Similar GDP per capita and population density Low predicted risk High Why? More complex road structure in Tamaulipas Business flow and commuting traffic with the United States (Texas) in Taumaulipas 21

22 density Risk characteristics two different cases Durango Chihuahua Let s have a closer look at two departments and see what our model says: Chihuahua High risk Durango Low risk Similar GDP per capita and population density Low predicted risk High Why? More complex road structure in Chihuahua Business flow and commuting traffic with the United States (Texas, New Mexico) in Chihuahua 22

23 Strategic expansion Average light yield per municipality Predicted risk Mexico City 23

24 Patent 24

25 The near future 25

26 Fleet Risk Fleet routes can be modelled to find the best route between points taking into account: Accident Risk Time of journey Labour costs Petrol cost Emissions The choice can be minimise overall costs! 26

27 Take a closer look 27

28 Severity Risk Map We are working intensively to determine a way to estimate Risk of collision with the probability curve of severity on each of the risk dimensions we are using. In a univariate way i.e. Severity Motor Risk Map In a dynamic way Frequency plane 28

29 Risk Map & Telematics Motor Risk Map Environment Risk Telematics Driver Risk 29

30 Swiss Re & Big Data 30

31 What Swiss Re does and offers Swiss Re Big Data Big data analysis is used to improve internal services, such as product innovation or more advanced claims management. Big data sets are processed to get insights into client's business and better meet client's needs. We are Smarter together Swiss Re can provide client services or customized product offerings based on what is learned from analysing Big Data sets and, therefore, support business growth or business retention. Swiss Re is actively working together with clients in order to identify business opportunities and discuss options on how to concretely implement resulting innovation and improvements along the client's value chain. 31

32 Bonus A perverted way of thinking Big Data 32

33 Explosion LNG carrier 33

34 Thank you very much Let's Brainstorm! 34

35 35

36 Legal notice 2015 Swiss Re. All rights reserved. You are not permitted to create any modifications or derivative works of this presentation or to use it for commercial or other public purposes without the prior written permission of Swiss Re. The information and opinions contained in the presentation are provided as at the date of the presentation and are subject to change without notice. Although the information used was taken from reliable sources, Swiss Re does not accept any responsibility for the accuracy or comprehensiveness of the details given. All liability for the accuracy and completeness thereof or for any damage or loss resulting from the use of the information contained in this presentation is expressly excluded. Under no circumstances shall Swiss Re or its Group companies be liable for any financial or consequential loss relating to this presentation. 36