Sunday, August 19, 2018

Decentralised insurance model


29 July 2018
Decentralised insurance model 


Abstract

With the development of sharing economy and decentralised platforms like Bitcoin and Ethereum the financial products decentralisation has been the focal point for some projects and studies in the blockchain world. In this paper, we build a decentralised insurance model based on our researches and experience in REGA Risk Sharing and evaluate the model parameters to prove that the decentralised insurance can be an alternative solution to the traditional insurance approach.
Keywords

Decentralised insurance, blockchain, smart contract, crowdsurance, valuation, marketing.

1. Introduction

The decentralised insurance or crowdsurance meaning people unite in communities to provide a guarantee of compensation for unexpected loss. Compared to traditional insurance, in crowdsurance there are no insurers, intermediaries and brokers, all the processes being controlled and managed by programs stored in blockchain and executed by a distributed virtual computer. The main question in this case is how to motivate people inside the community to use crowdsurance programs in the right way to provide the risk coverage and fight the fraud. Some people intensions and actions can be in opposition of these objectives but in the decentralised economy if most actors are motivated properly to bring value to the community then the outcome of the whole process will be inline with common goal. 
There are two groups of people using crowdsurance program: A - people who join crowdsurance pool to get risk coverage and B - people who help the first group manage risks inside the crowdsurance pools. We will call the first group a crowdsurance community or community and the second one - experts. To join the crowdsurance community someone should buy a crowdsurance token (CST) and transfer a crowdsurance join amount to the crowdsurance token smart contract. To became an expert a person should buy a license, Risk Sharing Token (RST) and transfer an expert join amount to the RST smart contract. Funds collected in RST smart contract are using for crowdsurance product development and for the initial crowdsurance pools capitalisation. Inside the crowdsurance token smart contract there is three level pool structure: super pool, pools and sub pools. Each CST belongs to one sub pool that belongs to one pool and all pools included in super pool. The crowdsurance join amount is divided between different levels in there following way: 50% goes to sub pool, 20% to pool and 10% is transferred to super pool. The rest 20% is crowdsurance smart contract commission. A CST holder can activate the token to initiate the risk coverage period during which he/her can submit a claim to receive crowdsurance payment in case of unexpected event. The submitted claim will be processed by the crowdsurance program and will be submitted to an expert jury for final approval. For each outstanding claim the randomly selected jury will be created and experts will vote for the case using crowdsurance token smart contract. During the claim voting period (up to 48 hours) any jury member can review claim supported documents and vote in favour of payment or against it. The claim payment will be processed by the crowdsurance smart contract based on voting result and the CST holder could receive the claim payment if payment was approved by the jury. A part of the smart contract commission (3%) will go to the experts to compensate transaction fee and working time. If an expert was selected to vote but did not cast the vote during the voting period then he/her will not receive the compensation. Otherwise all other experts will receive expert commission even if they did not selected to vote. The crowdsurance smart token linked to the Risk Sharing Token through the super pool. The RST value will rise if the super pool balance will be more then a certain threshold at the end of risk coverage period as result of the experts performance. In the next section we will describe the crowdsurance model that explains  an expert motivation to process claims and vote for the legitimate cases and decline the fraudulent ones. The model takes into account a community member satisfaction level for different voting outcomes.
2. The model

The model consists of crowdsurance token smart contract working through two risk coverage periods. Similar to [1] community member decision making to join crowdsurance pool is modelled using the the surplus the member obtain from the decentralised product experience. The surplus is the difference between a member’s valuation for the product and the amount he/she must transfer to join it. It’s assumed that a member joins the crowdsurance pool, if the offer yields a nonnegative surplus. Let’s assume that pastedGraphic.png is the average risk coverage period for the community member that did not submit any claim. For a member who has submitted the claim in the first risk coverage period and has received the payment the product valuation for the next period should be more that the average one. From the other hand if claim was rejected then the valuation must be less. So, we can assume that for the member that has received the claim payment the crowdsurance contract lasts pastedGraphic_1.png and the rejected claim results in pastedGraphic_2.png periods. If pastedGraphic_3.png is group of members who has submitted the claim then the super pool financial result after the two risk coverage period can be calculated by the following expression: 
pastedGraphic_4.png                                                (1)
where pastedGraphic_5.png is crowdsurance join amount and pastedGraphic_6.png is the average super pool balance update due to the claim processing. To calculate pastedGraphic_6.pngwe need to consider the following cases:
  • the submitted claim pastedGraphic_7.pngis legitimate one;
  • the submitted claim pastedGraphic_8.pngis fraudulent one.
In both cases we will have the following outcomes for the expert voting procedure:
  • pastedGraphic_9.png the claim is approved by the jury and member will receive claim payment amount pastedGraphic_9.png;
  • pastedGraphic_5.pngthe voting procedure has been finished by timeout and the number of collected votes is less than specified limit or the number of positive votes is equal of number of negative votes. In this case the member will receive join amount pastedGraphic_5.png;
  • pastedGraphic_10.pngthe claim has beed declined by the jury and the member not receive any payment from the crowdsurance pool. 

Based on assumption that the member product valuation will change he/her average risk coverage period we can say that for pastedGraphic_11.png voting outcome will credit the super pool balance to pastedGraphic_12.png and the pastedGraphic_13.png will debit it pastedGraphic_14.png . We also can assume that pastedGraphic_15.png will bring the member product valuation surplus to 0 but it should not change the average crowdsurance period. Please note, that we are considering the crowdsurance model where all legitimate claims can be paid out using sub pool and pool funds and pastedGraphic_11.png, pastedGraphic_15.png voting outcomes are not using the super pool balance to pay claim in full pastedGraphic_9.pngor to return crowdsurance join amount pastedGraphic_5.png. On another hand fraudulent claims can have negative impact on super pool balance due to the shortage of funds in corresponding sub pool and pool. Thereby pastedGraphic_16.png outcome will change the super pool balance to pastedGraphic_17.pngand pastedGraphic_18.pngto pastedGraphic_19.pngwhere pastedGraphic_20.pngis probability of the fund shortage in sub pool and pool. The following table summarise described above outcomes:
Claim types / Voting results
D
J
Z




Legitimate - L
0.1Jx
0
-0.1Jy




Fraudulent - F
0.1Jx-Dp
-Jp
0




We suppose that for member submitted fraudulent claim voting results pastedGraphic_18.pngand pastedGraphic_21.png does not change the average crowdsurance period pastedGraphic.pngon another hand if fraudulent case has been paid then the member will more probably join the next period to try cheat again.
Now we can calculate pastedGraphic_6.pngusing the following expression:
pastedGraphic_22.png
Applying conditional probability theory, the above equation can be rewritten as: 
pastedGraphic_23.png
pastedGraphic_24.png 
pastedGraphic_25.png
pastedGraphic_26.png
where  pastedGraphic_27.png
We should assume that number of fraudulent claims is much less then the number of legitimate one, so we are adding the following condition: pastedGraphic_28.pngto the expression 2 where pastedGraphic_29.png. With this assumption we can try to calculate the probability of the funds shortage pastedGraphic_20.png. If N is total number of claims, NL is number of legitimate pastedGraphic_30.pngclaims and NF is fraudulent ones then pastedGraphic_31.png and pastedGraphic_32.png. In this case we can estimate pastedGraphic_33.pngwith the following expression:
Suppose we a have tree level pool structure where K is number of sub pools andpastedGraphic_34.pngis claims distribution between sub pools. For any distribution we have: pastedGraphic_35.pngand if there is pastedGraphic_36.pngfor pastedGraphic_37.pngthen we have the following condition pastedGraphic_38.png. Thereby for any sub pool claim distribution we can have only one pastedGraphic_36.png. In our model we еhave assumed that if for some claim distribution pastedGraphic_39.pngis true for all j then all claims from this distribution can be paid out using sub pool and pool funds without affecting the super pool balance. Thus to calculate the probability of event when claim payment affects super pool balance we need to calculate the number of claim distributions with one member pastedGraphic_40.png. For such member we can have the following set of values pastedGraphic_41.pngand for any pastedGraphic_42.png where pastedGraphic_43.pngwe can have pastedGraphic_44.pngdistributions . So the total number of distributions is pastedGraphic_45.pngand pastedGraphic_46.png         (3)

3. Model parameters valuation

Now we need to check that it’s possible to find pastedGraphic_27.pngwhere pastedGraphic_47.png. Let’s consider the following expression pastedGraphic_48.png where pastedGraphic_49.png and pastedGraphic_50.pngwhere pastedGraphic_51.png. If pastedGraphic_52.png then pastedGraphic_53.pngand if pastedGraphic_54.pngthen pastedGraphic_55.png
Consider the following outcome probabilities example

In this case pastedGraphic_56.pngand if pastedGraphic_57.pngthen pastedGraphic_58.png
Suppose the total number of claims is 100 and number of sub pools 5 then pastedGraphic_59.pngand pastedGraphic_60.png
Let’s return to the definition of pastedGraphic_6.png. Now we can see that pastedGraphic_47.pngif pastedGraphic_61.pngor pastedGraphic_62.png. In our example pastedGraphic_63.pngand pastedGraphic_64.png . Thus pastedGraphic_47.pngif pastedGraphic_65.png
If pastedGraphic_66.pngthen we can assume that pastedGraphic_67.pngand all member who has submitted legitimate claim and did not received the payment left the crowdsurance pool after the first period. On another hand we can assume that all members that have received the approval have two periods and pastedGraphic_68.png. Based on our example outcome probabilities we can see that for pastedGraphic_68.pngand pastedGraphic_67.png pastedGraphic_47.png. Please note then the minimum value for pastedGraphic_69.pngin our example that gives pastedGraphic_70.pngis pastedGraphic_60.pngand for pastedGraphic_67.pngwe need pastedGraphic_71.pngorpastedGraphic_72.png.
4. Summary and conclusion

In the previous sections we have build crowdsurance model with expert performance function (2) that explain expert motivation to consider claims properly voting to legitimate ones and reject fraudulent attempts. Based on expression (2) we can see that if experts have decided to vote again all cases including legitimate ones then the value of expert performance function pastedGraphic_6.pngwill become negative that brings decrease to RST token value. On another hand if experts have decided to approve all cases including fraudulent ones then this decision will change the risk balance in next crowdsurance period due to the fact that more fraud attempts will appear and the probability pastedGraphic_20.png of pool / sub pool funds shortage will increase. Thus with increase of pastedGraphic_20.pngand pastedGraphic_73.pngthe value of pastedGraphic_25.png in (2) can became more substantial and again the expert performance will became negative. 
We also have evaluated the model parameters pastedGraphic_27.png using example for voting outcome distribution and have proved that there is not empty set of outcome distributions that gives realistic value to the model parameters pastedGraphic_27.png. It’s save to assume that the average crowdsurance period will be around 1.1 and in our example distribution we check that for pastedGraphic_67.pngthe value for expert performance function pastedGraphic_6.pngwill be positive for all pastedGraphic_69.pngfrom pastedGraphic_74.png.
Based on the above summary we can conclude that the decentralised insurance product with two group of people with different motivations connected thought blockchain program can be created and function to bring value to both groups. We hope that it will help crowdsurance to became an alternative solution for risk coverage and will change insurance landscape in the nearest future. 
References

[1] Amir Gandomi and Saeed Zolfaghari, 2011 “Profitability of Loyalty Programs in th Presence of Uncertainty in Customers’ Valuations” Proceedings of the 2011 Industrial Engineering Research Conference