Deep Representation Learning using Stacked Autoencoder for ...€¦ · A novel approach for loss...
Transcript of Deep Representation Learning using Stacked Autoencoder for ...€¦ · A novel approach for loss...
Deep Representation Learning using Stacked Autoencoder for General Insurance Loss Reserving
1
Insurance Data Science Conference ETH Zurich14th June 2019
Satya Sai MudigondaSenior Tech Actuarial Consultant
+91 - 9603573032
Phani Krishna KandalaPricing Actuary, Swiss Re
+91 - 9182472136
Agenda
Introduction
Key Concepts: Reserving & Neural Networks
Motivation
Experimental Setup
Results
Next Steps
Q & A
2
Introduction
3
A novel approach for loss reserving based on blended unsupervised and supervised deep neural networks.
Non-linear latent features or deep representation extraction
Use the extracted latent features from step 1 as a starting point of multilayer perceptron
Estimate of Loss reserves
Auto Encoders Neural Network Output
UNSUPERVISED LEARNING SUPERVISED LEARNING
STEP 1 STEP 2INPUT
FEATURE 1
FEATURE 2
FEATURE 3
FEATURE N
Reserving – Key Concepts
To determine liabilities to be shown in accounts
To check adequacy of reserves
To evaluate business performance
To measure profitability of business currently being written
To calculate loss ratios and combined ratios
To perform insurance valuation
4
Development Year
0 1 2 3 4 5 6 7
Cla
ims
Occ
urr
ing
Year
2005
2006
2007
2008
2009
2010
2011
2012
TRAIN DATA
PREDICTION
Neural Network – Key Concepts
5
INPUT LAYER• Inputs
information for the neural network to process
• Each circle represents one feature
HIDDEN LAYER• These layers do all the processing for neural networks• Higher hidden layers higher accuracy of neural network• Each layer consists of nodes that mimic our brains’ neurons• Nodes receive information from the previous layer’s node• Nodes are multiplied with weights and bias is added to it
OUTPUT LAYER• Brings together
the information from the hidden layer
• Contains all the information needed for the program
INPUT LAYER
HIDDEN LAYER 1 HIDDEN LAYER 2
OUTPUT LAYER
Motivation of the paper
Ability to calculate Loss Reserves
6
• Auto Encoders
• Stacked Auto Encoders
Using Current applications of
Without choosing a specific reserving method
For any class of business
Independent of length of the Tail.
Auto Encoders - Overview
7
x2
x3
x1
x4
ො𝑥2
ො𝑥3
ො𝑥1
ො𝑥4
h1
h2
ENCODING DECODING
h = f(x)HIDDEN LAYER
INPUT LAYER ො𝑥 = g(h)OUTPUT LAYER
Stacked Auto Encoders - Overview
8
x1
x1
x1
x1
x1
x1
+1
ℎ12
ℎ22
ℎ32
+1
ℎ11
ℎ21
ℎ31
ℎ41
+1
ො𝑥1
ො𝑥1
ො𝑥1
ො𝑥1
ො𝑥1
ො𝑥1
ℎ11
ℎ21
ℎ31
ℎ41
+1
ℎ12
ℎ22
ℎ32
ℎ42
ℎ12
ℎ22
ℎ32
+1
P(y=0 |x)
P(y=1 |x)
P(y=2 |x)
INPUT
FEATURES IOUTPUT
INPUT(FEATURES I)
(FEATURES II)
OUTPUT
INPUT(FEATURES II)
SOFTMAXCLASSIFIER
Key Research Contribution from the Past
Neural Network embedding of the ODP
reserving model -Mario Wuthrich1
Insights from Inside Neutral Networks:
Switzerland Actuaries Association2
9
1. Gabrielli, A., Richman, R., W¨uthrich, M.V. (2018). Neural network embedding of the overdispersed Poisson reserving model. SSRN Preprint
2. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3226852
Experimental Setup
10
Data: Data has been taken from Bjorn Weindorfer1 paper
1 - https://www.uio.no/studier/emner/matnat/math/STK4540/h18/course-material/chainladder.pdf
Loss Reserve Predictions
11
Three different approaches have been compared with respect to Loss Reserve Predictions
PREDICTED LOSS RESERVES
Auto Encoders Neural Network Result 2
INPUT
ACCIDENT YEAR
DEVELOPMENT YEAR
Neural Network
Stacked Auto Encoders
Neural Network Result 3
Result 1
True Reserve Predicted Reserve Bias Bias Percentage
17352 27365.64 10013.64 57.71
12
Reserve Prediction with Neural NetworkResult 1
Accident Year
Res
erve
Val
ue
13
Reserve Prediction with Auto-Encoder + Neural Network
Result 2
True Reserve Predicted Reserve Bias Bias Percentage
17352 13750.39 -3601.61 20.75
Accident Year
Res
erve
Val
ue
14
Reserve Prediction with Stacked Auto-Encoder + Neural Network
Result 3
True Reserve Predicted Reserve Bias Bias Percentage
17352 14737.26 -2614.74 15.06
Accident Year
Res
erve
Val
ue
Potential Benefits to Insurer
Fastens the process of prediction by identifying the most important features
Useful to learn the 2 way and 3 way interactions in the claims data. For example: Claim type and cause of claim
Highlights the main reasons for claims reserves increase/ decrease
15
Next Steps
Reduction in bias by optimising the Neural network and Autoencoders.
Working on different LoBs without any embedding function(s)
Working on individual claims data to automatically highlight the predominant features increasing/decreasing the reserves.
• Adequate capital allocation.
• Suitable Reinsurance product(s) selection.
• Efficient Risk Management on insurance risk front.
16
17
Q & A
• What is novel about this idea?• What functions of insurance company will be able to use this idea?• What are the benefits to insurance companies?• Are you using any proprietary software/packages?• Is this approach used in any other industry?• How this approach can be improved further?• What are the limitations of this approach?• What reserve prediction methods does this idea support?• How this approach is superior to using Neural Networks on stand
alone basis?
Acknowledgements
• SSSIHL for providing Technology Labs and Actuarial Support
• Dr. Pallav Kumar Baruah, Department of Mathematics and Computer Science
• Arun Kumar K
• Nikhil Rai, M.Tech
• Rohan Yashraj Gupta, Actuarial Research Scholar
18