The live event ticketing industry has long struggled with fraud, scalping, and overpriced tickets. These issues are becoming increasingly prominent, with Live Nation, the dominant market leader and owner of Ticketmaster, recently facing antitrust concerns from US lawmakers.

DLT offers a timely solution: a decentralized architecture could address systemic problems and consumer frustrations, and create a win-win for all market participants. It also opens the door to more refined approaches to customer loyalty in the events space.

• Immutable Digital Ticketing: Blockchain-based tickets can no longer be duplicated or altered nefariously, with their unique digital signatures rendering fraud and counterfeits virtually impossible. As digital assets, tickets are easily tracked on a system that natively supports issuance, redemption, and accounting workflows.
• Visibility on Data: Digital ticketing provides event organizers with more granular consumer data. All transactions and ticket transfers, including previously invisible secondary sales, are recorded. This offers a bird’s-eye view of the entire ticketing ecosystem: who is buying tickets, how they are buying them, and how they are using them. It also presents opportunities for new, innovative loyalty initiatives powered by a deeper understanding of customers.
• End Scalping and Ticket Hoarding: The requirement for credentials at the protocol level provides one solution to the problem of bad actors. Alternatively, as the technology emerges in the public blockchain space, decentralized or sovereign identity could play its own role. Whichever strategy is seized upon, by removing intermediaries, predatory resellers, and bots, a decentralized architecture would help organizers clean up peer-to-peer ticket sales. They could then place limits on the markup price of resold tickets or require that all resale transactions take place on an official platform.
• Smart Contracts: Sophisticated code can establish a fairer, more transparent playing field for all participants. Smart contracts could help to render platforms regulatorily compliant, in turn, lowering the legal and regulatory burden for participating organizations. Automation could even enforce ticketing rules and enhance user experiences, while reducing operational costs for organizers. Limits could be placed on the number of tickets an individual could purchase and refunds could be provided to ticket buyers the moment a planned event is canceled.
• Novel Customer Experiences: The potential for improvements in the customer experience are already being demonstrated by current experimental use cases, from Ticketmaster enabling event organizers to issue NFTs to the Dallas Mavericks issuing NFTs as collectible rewards for loyal game attendees.

It’s time for the ticketing industry to evolve. A decentralized architecture would create the ability to uproot bad actors and create a more data-enabled, secure, transparent, and equitable system for all parties.

This requires some creative thinking from relevant innovators. One path would be to integrate a decentralized ticketing architecture with a modern tokenized loyalty system, with the aim of maximizing the crossover benefits. If we take, for example, Belgium’s Kate Coin — this could present exactly the sort of synergy needed to generate lift.

For both blockchain loyalty and blockchain-based tickets, DLT offers improved transparency, security, efficiency, and flexibility. Both use cases also share the highly disruptive quality of programmability. Tokenized loyalty points can be calibrated to serve only specific groups of customers, with conditions placed on issuance, redemption, and exchange options. While the programmability afforded by DLT to ticketing, combined with NFTs, for example, opens up avenues for more dynamic, creative experiences and collaborations between organizers and artists/celebrities.

Blockchain’s disintermediation will also have a profound impact on both use cases. Removing intermediaries from the ticketing industry can reclaim the space for organizers and consumers, generating engagement and eliminating pain points like excessive pricing. And for loyalty, it can lead to smoother, more immediate redemption options, shrinking the ‘earn and burn’ points cycle.

For all of the above reasons, combining ticketing and loyalty through blockchain has the potential to create an entirely new, dynamic, peer-to-peer market for experiences.

IntellectEU specializes in co-creating enterprise-grade blockchain solutions, helping leading institutions navigate complex new domains through our deep business and technical expertise. If you would like to discuss this or any other potential use cases, reach out to us at info@intellecteu.com.

Blockchain Is The ‘Golden Ticket’ the Live Events Industry Has Been Waiting For was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

This article consists of 3 main sections. First, we explain option contracts and possible ways to determine their price. Next, we discuss the quantum amplitude estimation algorithm and the main building blocks which constitute it. Finally, we introduce 5 recent and improved variants of this algorithm and benchmark their performance and their quantum hardware requirements.

Option contracts basics

Derivative contracts constitute one of the main classes of financial assets. They derive their value from one or more underlying assets (hence the name) like stocks or commodities and are commonly used as a crucial part of hedging strategies or for speculation purposes. There are several different types of derivatives like futures or swaps, but here we will only talk about options. In their simplest form, the buyer of an option contract acquires the right, but not the obligation, to buy (this is known as a call option) or sell (this is known as a put option) the underlying asset for a predetermined price (the strike price) at a predetermined date (the expiration date). This means that, for the case of a call option, if the price of the asset at the expiration date is above the strike price, the option contract can be exercised to buy the underlying asset for the strike price and then immediately sell it for the higher market price and as such lock in a profit. If however the asset price at expiration is below the strike price then the option contract simply isn’t exercised. For put options, on the other hand, it’s the other way around. That is, a profit can be made if the underlying asset price is below the strike price by using the option contract to sell the asset for the strike price and then immediately buying it for the lower market price, whereas the option contract isn’t exercised if the asset price is above the strike price.

Determining a fair price

The above discussion could give the impression that entering an option contract offers the opportunity to realize risk-free profits. As we all know however, in reality there is no such thing as a free lunch. In the case of options trading this reality is reflected in the fact that a premium needs to be paid in order to enter an option contract. Therefore the challenge which option traders face (whether they are retail or institutional) is to estimate whether or not the potential profit which could be achieved by exercising the option will outweigh the premium which has to be paid to enter the contract. For the simple option types discussed in the above paragraph, which are known as European options, there exists a closed analytical formula to determine the expected value O of the option. This is known as the Black-Scholes formula and for a call option it is given by

with K and T the strike price and time to maturity of the option contract, with S₀, r, and σ the spot price, drift (assumed to be equal to the risk-free market rate), and volatility of the underlying asset, with N the standard normal cumulative distribution, and with

In principle this value O first needs to be discounted, taking into account the risk-free rate, before it can be compared with the option’s current premium. However, since this is not a computationally challenging process (given current and future interest rates), we will not do so in this article and only focus on the expected option value at expiration. It should be noted at this point that several significant assumptions need to be made in order to arrive at the above expression. Some notable examples include the absence of arbitrage opportunities and dividends, but perhaps most importantly the assumption that the price of the underlying asset follows a geometric Brownian motion with a constant drift (the return of the asset) and volatility, which implies that the asset price follows a log-normal distribution at any given time. These assumptions limit the accuracy and applicability of the Black-Scholes formula and it means that if we want to model options without these assumptions, or if we want to model more exotic options like Asian options or barrier options, we need to resort to numerical methods. One of the most widely used numerical methods in this context is the Monte Carlo method, where the general idea is to generate a large number of possible price paths for the underlying asset, calculating the option value for each of them, and then averaging across all the possible paths to obtain the expected option value. A simplified, very small-scale example of this method is shown here:

The estimation error ε of this approach scales as

with M the number of samples, and as a result large financial institutions typically run large-scale simulations overnight in order to achieve the desirable accuracy for large portfolios of complicated options.

Taking the options to quantum town

The scaling of the Monte Carlo method and the associated computational cost put serious restrictions on the size and complexity of the portfolios which can be numerically simulated and/or on the accuracy of those simulations. This is because it’s very important to avoid that the results of a simulation are already irrelevant or outdated by the time the calculations are completed, for example because the markets have opened again. This is where quantum computing could offer a very significant advantage over classical methods because, by means of a quantum algorithm known as quantum amplitude estimation (QAE), quantum computers can perform Monte Carlo simulations for which the estimation error scales as 1/M, i.e. quadratically faster than classical Monte Carlo simulations. We will now discuss the three main parts which constitute this algorithm.

Loading the asset price probability distribution

The first step of the amplitude estimation algorithm consists of loading the desired probability distribution for the price of the underlying asset into a quantum register of n qubits. In mathematical terms, we need to construct an operator P which does the following:

i.e. it maps the all-zeroes initial state to a superposition of the possible spot prices of the underlying asset (represented by the integers xᵢ in the range [0, 2ⁿ-1]), with the weights given by the square root of the corresponding spot price probabilities pᵢ. There are several ways to construct this operator. A first way to do this is by using the generic algorithm which was proposed in this paper, and which is implemented for example in the initialize method of Qiskit. However, this method can yield sub-optimal quantum circuits which in general can have a depth which scales exponentially with the number of qubits, making it unsuitable when applying it to real-life problems. A second way to load a probability distribution is by using the Grover-Rudolph method which allows for efficient state preparation if the probability density function is efficiently integrable. The problem with this method, however, is that it already requires classical Monte Carlo samples itself, making it useless as a subroutine for quantum Monte Carlo simulation because it prohibits quantum advantage. Another way to load a probability distribution is by making use of a quantum generative adversarial network (qGAN), which makes use of a quantum generator (i.e. a variational circuit) and a classical discriminator (i.e. a neural network). This allows for the learning and approximate loading of generic probability distributions by feeding data samples to the qGAN. The upside of this hybrid quantum-classical approach is that it leads to very shallow quantum circuits (making it very suitable for NISQ (noisy intermediate scale quantum) devices), while the downside is that the efficiency and overhead of the classical training process are unclear. New and improved methods for loading data into a quantum register are continuously being developed, but for the small-scale demonstrations in this article we will stick to the generic approach discussed at the start of this paragraph.

Constructing the option payoff and Grover operators

The second step of QAE consists of constructing an operator F which realizes the following transformation:

with f(x) the function defining the option payoff as a function of the underlying asset price, e.g. max(x-K,0) for a European call option with strike price K. This can approximately be achieved by means of elementary quantum operations and some ancilla qubits. Based on the above equation, we can see that we are interested in determining the total probability of obtaining the state 1 when measuring the last qubit, because it is given by

which is exactly the option payoff, averaged across the loaded probability distribution, i.e. it is equal to the expected value O of the option which we were looking to find. Note that at this point we could simply measure the last qubit and as such obtain an estimate for P₁, however this approach does not scale better than the classical approach as we will later illustrate. QAE, on the other hand, allows the efficient, quadratically faster than classical, estimation of P₁. For the reader with technical quantum computing knowledge (the non-technical reader needn’t worry and can simply skip to the next subsection): QAE is able to achieve this speedup by means of the Grover operator

with Iᵤ the identity operator of dimension u, Z the Pauli-Z operator, and A=F(P⊗I₂). The Grover operator Q has two highly degenerate eigenvalues ±θ which are mapped to the desired probability according to P₁=sin(θ/2)².

Calculating the eigenvalues

The third and final step of QAE consists of actually determining the eigenvalues of Q. The original, canonical version of this algorithm leverages quantum phase estimation to accomplish this goal. However, this requires many sequential controlled applications of Q, and as such leads to significant circuit depths, making it unsuitable for current NISQ devices. This is why recently several different variants of QAE have been proposed that do not use quantum phase estimation, but still provide a quadratic speedup over classical methods. These include maximum likelihood QAE, simplified QAE, iterative QAE, faster QAE, and power law QAE. The general idea behind all of these is that they measure QᵏA|0> for cleverly selected different values of k, such that combining the results for the different values of k leads to a quadratically improved estimate of P₁. The fact that any quantum circuit in these algorithms only involves a single power of a non-controlled version of Q makes them much more suitable for current NISQ devices.

Algorithm benchmarking

For all of the following results, we consider a single European call option with a strike price of 1.9, a time to expiration of 300 days, and an underlying asset with S₀=2, r=0.04, and σ=0.1.

Comparing the QAE variants

We will now investigate how all of the QAE variants listed in the previous section compare when we apply them to the problem of option pricing. To do this we check how the estimation errors of these algorithms scale as a function of the number of samples used:

We can see that all of the QAE variants outperform classical Monte Carlo simulation above some threshold number of samples, at which point they achieve better accuracy for the same number of samples or, in other words, require less samples to achieve the same accuracy. Furthermore, all of the results show a linear dependence in this log-log plot. This means that the estimation error ε indeed decays according to a power law with increasing number of samples M, i.e. ε=aMᵇ. The values of a and b for each QAE variant can easily be determined by fitting the results, which gives:

Note that we did not include results for the “simplified” method because it has a very large constant factor, making it slow compared to the other methods and therefore impractical for real-life applications. Looking at the exponents (i.e. the values for b) we can see that, as expected, the error of the classical method scales as the inverse of the square root of the number of samples. Furthermore, the naive quantum method scales the same way (i.e. it has the same value for b) as the classical method, meaning that it does not provide any speedup. Turning to the QAE variants, we see that, apart from numerical deviations (especially for the “likelihood” and “faster” methods) as a result of the stochastic nature of the algorithms and the limited number of simulation repetitions, the exponents are indeed close to -1, which goes to show that these algorithms indeed scale quadratically better than classical (and naive quantum) methods. Taking into account the constant factors (i.e. the values for a) as well, it turns out that in practice the “iterative” and “likelihood” methods perform better than the “canonical” and “faster” methods. The “power” method is a special case because it features a parameter β which interpolates between QAE-like and classical error scaling, i.e. when it’s closer to 0 the algorithm achieves better error scaling (as can be seen in the above table). However, this comes at the expense of an increase in required quantum hardware resources.

Apart from the error scaling of the different methods, it’s also paramount, in the context of the current NISQ devices, to compare the required circuit depths and number of two-qubit gates (i.e. CNOT gates) to achieve a certain accuracy (for this comparative table we took ε≈0.02) for each of the methods. We did this by transpiling all of the quantum circuits using the native gate set of the IBMQ-Montreal device and obtained the following results:

This shows that the canonical QAE algorithm requires very deep quantum circuits containing a lot of two-qubit gates (in the order of millions), confirming the doubts which we had already raised at the end of the previous section. The QAE-variants, on the other hand, allow for much shallower quantum circuits containing much less two-qubit gates, with the “iterative” method again emerging as the best performing one. It should be noted that the “power” method can yield shallower circuits (for sufficiently high β values), however as we have seen in the previous table this comes at the expense of less efficient error scaling. The same holds true for the “naive” method, which has by far the shallowest circuits but at the same times doesn’t offer any advantage over classical methods.

Performance on “real” quantum hardware

Considering the above results, it’s worth investigating whether the “iterative” method can be used to obtain accurate, better-than-classical results when it is run on a real quantum computer. Up to now we have been using Qiskit’s qasm simulator, which mimics a perfect, so-called fault-tolerant quantum computer. However, in reality current quantum computers are imperfect and introduce many different errors. That is why we also ran the “iterative” method on Qiskit’s fake Montreal and fake Guadalupe simulators, which mimic the corresponding real quantum devices using system snapshots. Doing this for the settings of the first data point of the “iterative” method in the previous figure (leading to circuits depths and CNOT counts of the order of several hundreds) gives:

The results show that, as expected, the accuracy of the results of the qasm simulator is very good and doesn’t deteriorate with increasing qubit count. When using a simulator mimicking a real quantum computer, however, the results are less accurate than the qasm simulator results and do deteriorate with increasing qubit count. The Montreal machine is able to achieve more accurate results than the Guadalupe machine thanks to its lower error rates, although they may still not be sufficiently accurate for real-life use cases. However, it should be noted that we did not employ any form of error mitigation techniques, which would help to increase the accuracy of the results. Furthermore, the error rates of these quantum computers are expected to improve in the future, as such improving the accuracy of the results even further, especially when they reach the scale at which quantum error correction can be implemented.

In conclusion, we have compared the performance of different QAE variants in the context of option pricing and empirically found that all of them offer the same quadratic speedup (i.e. quantum advantage) over classical Monte Carlo methods as the canonical QAE algorithm. However, the requirements in terms of quantum hardware of these variants are much less stringent, opening up the door towards near-term real-life use cases leveraging NISQ devices. Finally, note that even though this article focused on European call options, the same quantum algorithms and techniques can also be (and we in fact have) implemented to tackle the pricing of other and more exotic options, the pricing of other derivative types like futures and swaps, and even more general risk analysis.

Do you have additional questions, or do you want to have further discussions in a business and/or scientific context? Don’t hesitate to reach out to Matthias Van der Donck at matthias.vanderdonck@intellecteu.com

Quadratically improved option pricing using quantum computing was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

IntellectEU is pleased to announce that payLOCATOR, IntellectEU’s innovative payment tracking and sharing solution, has received the SWIFT gpi for Corporates label for 2022. Technical and functional validation made by SWIFT experts shows that payLOCATOR has completed the gpi for Corporates integration and is capable of processing gpi for Corporates-related flows. Congratulations to the team!

Enabling direct access to payment information for banks’ clients and their counterparties

The GPI Tracker information is accessible primarily for Financial Institutions. However, corporations can get access to payment status tracking thanks to payLOCATOR. What’s more, they can share these payment statuses with counterparties to improve their cash management and operational efficiency and avoid a lack of awareness of charges and possible delays, difficulties with forecasting, lack of confidence in the fulfillment of obligations by a counterparty, incomplete reporting and more.

Enter payLOCATOR

payLOCATOR tracks payments and reports on payment flows, commission charges, and currency exchange rates in one single user interface, easily accessible via API. It allows banks’ clients to share payment information from the system with their business partners via secure credentials. Using the dashboard, corporates connected to payLOCATOR can see in real-time how their payments move through the correspondent SWIFT banking chain, from inception right through until the beneficiary account is credited. And this goes for the banks’ corporate clients but also (if they receive the proper access) for their counterparties (suppliers, beneficiaries, …). Moreover, inbound payments can now also be tracked and shared, and receipt of payment can be more or less predicted, improving trust and transparency.

100% SWIFT certified

IntellectEU’s gpi-related initiative payLOCATOR makes sure that banks embracing SWIFT gpi can easily leverage this service for their Corporate client base. Users (=corporates) are able to achieve full transparency over where a payment is at any given moment, improving cross-border payments across the correspondent banking network. This is essential for corporates for whom speed, certainty, and a smooth international payments experience is an absolute must. -wait for update-

Give corporates easy access to payLOCATOR by leveraging SWIFT gpi

payLOCATOR’s payment tracking and data sharing features create a reduction in operating costs as a result of simplified payment investigation and dispute resolution, as well as optimized supply chain cash flows. IntellectEU is proud to support its clients — banks and technology vendors — to offer payLOCATOR benefits to their corporate clients, so they can remain innovative, customer-oriented leaders in the payments space.

About IntellectEU

IntellectEU is a global technology company focused on digital finance and emerging technologies. Since 2006, IntellectEU has developed its expertise in financial messaging and integration, being a SWIFT global premier service partner. IntellectEU continuously supports financial institutions and corporations adopting new solutions that face industry challenges. For more information about us and what we can do for you, please visit www.intellecteu.com or send us an email: info@intellecteu.com

IntellectEU’s payment tracking product payLOCATOR certified by SWIFT was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

Welcome back. The purpose of today’s blog post is to gain a better understanding of what payLOCATOR can do for your business from the perspective of the global corporate treasurer.

At the present moment SWIFT’s global payment initiative (gpi) ensures that international payments meet the industry’s need for speed, traceability and transparency. Through gpi, SWIFT and the global banking community have collaborated to put in place a solid standard for handling cross-border payments. It is used throughout the world, with over 11,000 customers in more than 200 countries, and counting.

To track and report on the speed and status of these global payments in real-time, IntellectEU has developed payLOCATOR — a tracking solution for global companies’ Treasurers. Today, corporates get payment information in a variety of formats from their banks, but payLOCATOR provides a holistic view of these payments in one format using a single window interface.

The only condition for a corporation to have access to the tracking information in payLOCATOR is that the enterprise’s servicing bank must be ready, both technically and operationally — to send its corporate client the payment status messages from the SWIFT gpi tracker via the SWIFT network. Consequently, in addition to being a member of the SWIFT gpi community, the servicing bank has also agreed for its client to receive a subscription to the g4c service (for more information on G4C, please read our previous blog post).

Then, payLOCATOR can be easily plugged into existing solutions via an API, meaning minimal implementation costs for the business.

Treasurers that need to report daily on myriad financial transactions in their industry and across multiple accounts, will find payLOCATOR especially useful. It is an excellent time-saving tool in that it consolidates all of the multi-bank payments into a single intuitive user interface, with convenient reporting features.

Want to learn more about the advantages using payLOCATOR presents to Corporate Treasurers ?

Book a meeting with our team here.

Oriana Claeys

Product Manager payLOCATOR

payLOCATOR. Main Concepts. Chapter 5 was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

Catalyst Blockchain Platform is now available on Corda, the leading distributed ledger technology platform from R3, created specifically for highly-regulated markets. Users benefit from both Corda Community Edition and Corda Enterprise Edition, using Catalyst’s highly-automated processes and intuitive user interface to streamline the development, deployment, and maintenance of their Corda networks and applications.

On Corda, privacy is paramount. The protocol enables scalable, secure data transactions between network participants while ensuring the highest level of privacy and security. Corda employs a unique peer-to-peer architecture that maximizes confidentiality, while Corda distributed applications offer exceptional flexibility and interoperability.

These features together make Corda ideal for deployment by financial institutions and financial service providers, governments, healthcare providers, insurers, and other regulated organizations.

“On Corda, Catalyst Blockchain Platform brings a whole new frontier of data privacy and ease of use to blockchain network development. It is especially relevant for those operating in challenging regulatory contexts, and we are extremely proud to be able to offer these organizations an easy pathway to blockchain adoption,” said Yana Koldra, Head of Product Management, IntellectEU.

Catalyst Blockchain Platform on Corda takes all of the strengths of the protocol and renders them easier and faster to work with by removing the technical barriers to entry. With Catalyst, users can build complex Corda infrastructure with just a few clicks, zero coding, and highly automated and optimized processes. To learn more about the protocol, visit www.catalyst.intellecteu.com/corda/.

Catalyst Blockchain Platform is developed by IntellectEU, a leader in distributed finance. Visit www.catalyst.intellecteu.com to learn more about the platform.

About IntellectEU

IntellectEU is a SWIFT partner and global leader in emerging technologies and digital finance. The company has a reputation for deep expertise in financial messaging and integration, however, in recent years IntellectEU has pushed the frontier of blockchain technology, becoming a founding member of the Linux Foundation’s Hyperledger in 2016. IntellectEU is a Certified Hyperledger Service Provider and the company has a partnership with leading protocol creator R3, retaining an experienced team of R3 Corda-certified developers. For more information about IntellectEU and what the company can do for you, please visit www.intellecteu.com

Catalyst Blockchain Platform launches on Corda was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

Summary

• Centralized databases are vulnerable to cyberattacks
• This has been demonstrated by an ever-growing list of high-profile cyber incidents and breaches
• In contrast, decentralized systems are innately robust, reducing cyber risks
• ClaimShare combines decentralization with confidential computing to deliver superior security in anti-fraud data-sharing — perfect for the insurance industry

A number of projects around the world have brought insurers together to combat insurance fraud. Typically, they establish a form of centralized anti-fraud database to facilitate improved inter-insurer data sharing within their ecosystem. This makes it possible to prevent potential crimes, such as duplicate claims fraud.

Unfortunately, while reducing their exposure to one potential crime, they render themselves more vulnerable to another: cyberattacks.

Cyberattacks are a fact of modern business. Each year, the percentage of companies reporting to have suffered at least one data security breach seems to rise. It has become clear that insurers are a prime target for potential cybercriminals, particularly as they store significant amounts of personal data.

• In 2021, the Asia Pacific component of global insurance giant AXA suffered a severe data breach resulting from a cyberattack. The incident led to 3TB of data loss, including identity documents, claims, reimbursements, and account and medical details.
• That same year, New Orleans-based Pan-American Life Insurance Group was hit by a cyberattack that crippled its communications. Details of the amount of data lost were not disclosed.
• And also in 2021, US-based health insurer Excellus Health Plan agreed to pay $5.1 million to settle with authorities over a 2013–2015 data breach that affected over 9.3 million people. This last case is an example of how cases can haunt insurers long after the initial incident is resolved.

There are countless cases.

Although centralized data is vulnerable data, databases continue to fulfill a vital function for organizations. And this can lead to problems, particularly for businesses operating in highly regulated markets.

Companies that handle the sensitive personal identifiable information of customers — such as insurers — need to ensure data is safeguarded, as compliance failings can be costly. Under the European Union’s GDPR, for instance, a data breach of sensitive information can lead to a fine of up to 4% of a company’s annual revenues (premiums). For this reason, it makes sense to invest in mitigation tools.

Fortunately, a solution exists, one that tackles both the issue of duplicate claims fraud and insurer data security.

ClaimShare is secure by design

Created by IntellectEU, experts in distributed finance and emerging technologies, ClaimShare makes it possible to compare claims data between insurers without exposing or storing personal identifiable information (PII) in a centralized location.

• ClaimShare is secured with R3 Corda Distributed Ledger Technology. The solution is decentralized, encrypted, and confidentially-focused, making it far safer than a centralized database from a data security perspective.
• ClaimShare uses confidential computing, preventing exposure of PII-data during the matching process.
• ClaimShare needs only a fraction (5–10%) of the PII-data to match claims (using fuzzy-matching), making it vastly more scalable than alternatives.
• ClaimShare does not store the personal data of the claimant or the insurer. This avoids any issues with compliance.

These features make ClaimShare the only practical solution to data sharing in the insurance industry. For more information, please find a short introduction to the technology here.

Reach out to us directly at claimshare@intellecteu.com or book a demo at www.claimshare.intellecteu.com.

Anti-fraud databases: the security challenges of conventional storage was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

Portfolio optimization

Portfolio optimization is the process of constructing a selection of financial assets, and assigning weights/amounts to them such that they optimize a certain objective function. This objective is usually a combination of minimizing volatility (i.e. risk) and maximizing returns, often subject to one or more constraints such as imposing a fixed total budget. There are many different ways to model this objective but one mathematical framework which stood the test of time is modern portfolio theory, also known as mean variance analysis. It was first introduced by Harry Markowitz in 1952 and its mathematical formulation is detailed in the methodology section at the bottom of this article.

Continuous vs discrete

There are two fundamentally different ways of tackling this problem, depending on whether the asset amounts are assumed to be continuous variables or discrete, integer-valued variables. The continuous approach in general permits finding the optimal solution efficiently, although it’s also susceptible to possible numerical instabilities depending on the data. However, these can easily be mitigated and as a result this continuous approach is adequate for retail investors. Institutional investors, on the other hand, typically trade in large, discrete blocks of assets, e.g. in ETF-creation and ETF-redemption packages. In order to find the optimal portfolio in this case, one could start from the solution obtained from the continuous approach and then round those amounts to the nearest allowed integer value for each of the blocks of assets. However, this leads to approximate results and a solution of the discrete approach is therefore expected to yield a more optimal portfolio. Considering the substantial budgets which are used by institutional investors, this means that a lot of money is potentially being left on the table when using the continuous approach.

Unfortunately, there are serious challenges associated with the discrete approach. Due to its non-convex nature and the fact that the number of possible solutions scales exponentially with the number of binary variables, it is part of the infamous class of NP-hard problems, but it is not part of the class of NP problems. This means that not only is it impossible to efficiently find the optimal solution, it is also impossible to efficiently determine whether a given solution is optimal or not.

Quantum computing to the rescue

The NISQ era

This is where quantum computers come into the picture. These are fundamentally new types of computers whose building blocks (the quantum bits or qubits) are governed by the quirky laws of quantum mechanics. As a result, completely new (quantum) algorithms can be developed which leverage the principles of quantum mechanics (e.g. superposition, entanglement, interference) and which can be executed on quantum computers in order to tackle problems which are intractable for classical computers. Although it is assumed (but not yet proven) that quantum computers will not be able to efficiently solve NP-hard problems, it is also believed (but again not yet proven) that they will be able to significantly speed up the process of finding an (approximately) optimal solution and/or to find solutions which are closer to the optimal solution.

However, the current quantum computers are not only limited in size (i.e. number of qubits), but also in quality, e.g. imperfect quantum gates (which are the fundamental building blocks of quantum algorithms) and limited qubit coherence times (the time span during which a qubit stays in a specific state). As a result, we cannot just run any quantum algorithm on these NISQ (noisy intermediate scale quantum) devices and expect to obtain useful results from them. Instead, we need to make sure that the total number of successive gates on each of the qubits (roughly equivalent to the so-called quantum depth) and the total runtime of the algorithm are under control.

A well-known quantum algorithm which ticks these boxes is the variational quantum eigensolver, or VQE. It starts by defining a quantum state which is determined by a set of parameters, and which is then used to calculate the value of the portfolio function (i.e. the equation at the top of the methodology section) with respect to this state. Thanks to the superposition principle, this quantum state can actually be a combination of multiple solutions. As a result, we can use a quantum computer to calculate the weighted average of the portfolio function with respect to the different solutions in the superposition. This is a quantum subroutine which is enclosed in an outer loop which classically optimizes the parameters of the quantum state in order to minimize the expectation value of the portfolio function. This is referred to as a hybrid quantum-classical algorithm.

State-of-the-art VQE variants

There are a few ways in which the standard VQE algorithm can be improved. We implemented and discuss three of them here. The first one is replacing the standard mean optimization with CVaR (conditional value at risk) optimization, which we will refer to as α-VQE. The idea is that instead of optimizing the weighted average of the portfolio function with respect to the solutions in the quantum state, we are actually more interested in optimizing the lowest-found portfolio function value among all of the solutions in the quantum state. However, simply using this lowest value leads to a non-smooth, ill-behaved objective function that is difficult to handle for classical optimizers. One way to avoid this issue while still keeping the emphasis on optimizing the lowest-found portfolio function value is to optimize the conditional value at risk, i.e. the average of the α% lowest portfolio function values found by the quantum computer. Note that the parameter alpha interpolates between the two extreme cases, i.e. α=100% corresponds to the average used in standard VQE and α=0% corresponds to the minimum value.

The second improvement of VQE which we implemented is filtering VQE, which we will refer to as f-VQE. This algorithm is a generalization of VQE and approximates the repeated action of a filtering operator on an initial quantum state by classically optimizing the parameters of this state. Applying the filtering operator increases the weights of the solutions in the superposition with low portfolio function values, and decreases the weights of the solutions with high portfolio function values. As a result, after each application of this operator the probability of finding the solution with the lowest possible portfolio function value increases.

The third and final way to improve VQE which we implemented is using a so-called warm start of the parameters, which we will denote with the suffix -w. We do this by solving the continuous version of the QUBO (defined in the methodology section at the bottom of this article), i.e. where the variables can take on any value in the range [0, 1]. As mentioned before, this type of continuous problem is generally efficiently solvable. For each variable this continuous solution is then transformed to a corresponding quantum state parameter such that for example a variable with a solution of 0.5 corresponds to an equal superposition of 0 and 1 for the corresponding qubit (more details can be found in the methodology section).

Algorithm benchmarking

For all of the following results, we use real stock data for the year 2020. We select assets randomly from a total set of 100 stocks in order to generate different instances of the portfolio optimization problem. Note that in our experiments the number of qubits required to solve the problem is equal to then number of assets. See the methodology section for further details.

Comparing the VQE variants

Now that we have a high-level understanding of these different VQE algorithms, we can have a closer look at how they compare when we apply them to the problem of portfolio optimization. To do this we check how the weight of the optimal solution inside the final quantum state changes as a function of the number of assets in our portfolio:

We can see that, in terms of this metric, the standard VQE algorithm is substantially outperformed by α-VQE, which in turn is substantially outperformed by f-VQE. At the same time, the results of both VQE and α-VQE are significantly improved by employing the warm start method. This is not the case for f-VQE, for the simple reason that there was little to no room for improvement. It’s important to note at this point that the optimal solution weight should not be interpreted as the success probability for finding the optimal solution. Each of these algorithms leads to a final quantum state which is a superposition of different solutions. Unfortunately, the laws of quantum mechanics do not allow us to see the full quantum state. Instead, we can only sample it and obtain the different solutions in its superposition with probability equal to the corresponding weight of the solution. If this probability is greater than or equal to the inverse of the number of samples (in our case 1024), we can expect to detect the corresponding solution. We were able to find the optimal solution for all VQE variants, all asset numbers, and for every problem instance, in the above figure. By extrapolating the above results to higher asset numbers, we see that standard VQE will be the most likely to fail to find the optimal solution, followed by α-VQE, and finally by f-VQE.

Quantum(-inspired) advantage?

The reason we could compare our results with a known optimal solution, is because for not too high asset numbers we can simply implement an exhaustive search of all the possible solutions, which is of course guaranteed to find the optimal solution. However, the calculation time for this brute force method scales exponentially, i.e. it is doubled every time an extra asset is added, and it is therefore not of practical use for real-life scenarios. If we compare the calculation time of the exhaustive search with that of f-VQE-w we obtain the following result:

For f-VQE-w we stopped the calculations when the lowest-value solution in the superposition has a weight which is at least twice the weight of the second-highest-weight solution, or if no new lowest value is found for 20 consecutive iterations. For all asset numbers in the above figure, f-VQE-w was able to find the optimal solution for each of the problem instances. We can therefore conclude that, for asset numbers greater than or equal to 22, f-VQE-w is a more suitable method than an exhaustive search.

However, because of its scaling problem, the exhaustive search is not applied in real-life scenarios and therefore it’s not really fair to use this classical method to compare our quantum(-inspired) results with. Instead, people often use heuristic algorithms which can quickly find good approximate solutions. Here we choose one of the most common heuristic algorithms for this type of problem, i.e. simulated annealing, to serve as the classical benchmark. Comparing the lowest value found by this algorithm with the lowest value found by α-VQE-w gives the following result:

Note that here we chose to use α-VQE-w instead of f-VQE-w because the former is significantly faster than the latter for higher asset numbers. Furthermore, since simulated annealing is in turn faster than α-VQE-w (obtaining an approximate solution takes 1–6 seconds for the above asset number range), we use the lowest value across 100 different runs to compare with our α-VQE-w result. That way, both methods are given more or less the same amount of time to return a solution. We can see that α-VQE-w outperforms simulated annealing, and that this difference between the two methods increases with increasing asset numbers.

There is one final but very important remark which should be made. All of the above results were obtained by using a variational quantum state which doesn’t contain any entanglement. As a result, these calculations can all be efficiently done on a classical computer, even for higher asset numbers. These types of algorithms are known as quantum-inspired algorithms, and even though they cannot lead to so-called quantum advantage, they can still provide improved results in terms of speed and/or quality. In order to find out whether these VQE variants have the potential to exhibit real quantum advantage in the future, we should start by adding entangling gates (i.e. CNOT gates) to the variational state and see whether this helps to improve the results:

The figure indicates that adding entangling gates can lead to improved results. This suggests that, when applied to problem instances of sufficiently large scale and executed on future fault-tolerant quantum computers, these VQE algorithms could allow to achieve quantum advantage for portfolio optimization.

In conclusion, we have compared the performance of different VQE variants in the context of portfolio optimization and found that the recent, state-of-the-art, improved versions can significantly outperform the standard VQE algorithm. Some of these algorithms could already achieve quantum-inspired advantage over standard classical algorithms today. Furthermore, when tackling a problem instance of sufficiently large scale on a fault-tolerant quantum computer, these algorithms could even achieve true quantum advantage in the future. Finally, note that even though this article discussed the specific case of portfolio optimization, the same conclusions apply to other combinatorial optimization problems such as arbitrage, transaction settlement, vehicle routing, …

Do you have additional questions, or do you want to have further discussions in a business and/or scientific context? Don’t hesitate to reach out to Matthias Van der Donck at matthias.vanderdonck@intellecteu.com

Methodology
The mathematical formulation of modern portfolio theory is given by:

In this equation, nᵢ, nᵢ,₀, pᵢ, and μᵢ are the amount, initial amount, price, and expected return of asset i, σᵢⱼ is the expected correlation between assets i and j, Nₐ is the total number of assets, B is the total budget, γ and ρ are parameters which control the risk aversion and the importance of the budget constraint, respectively, ν is the percentual transaction cost (note that the transaction cost in reality depends on the absolute value of the transaction amount, but for computational simplicity we use a quadratic dependence here, which has the same qualitative effect), and ν’ is a parameter which quantifies market impact.

The goal of portfolio optimization is to find the parameters nᵢ which minimize the above portfolio function H. The discrete version of this problem can be solved by transforming this equation to a QUBO by using binary encoding for the integer-valued variables, i.e.

where Nₚ is the number of bits of precision which are used for the integer variables, i.e. it defines the maximum value of nᵢ as nᵢ=2^Nₚ — 1. The total number of binary variables in this QUBO formulation (and thus the number of qubits required to solve it on a quantum computer) is given by N=Nₐ*Nₚ. Note that in the above form the nᵢ are all positive integers, meaning that we only consider long positions. However, this can easily be extended to allow for short positions by changing the above transformation such that it can also yield negative values. The QUBO formulation is in turn transformed to an Ising problem by promoting each binary variable to the Pauli Z quantum operator by means of the transformation xᵢ,ₛ=(1 - Zᵢ,ₛ)/2

For the expected returns μᵢ and expected correlations σᵢ,ⱼ we use the average values of the year 2020. For the prices pᵢ we use those for the last day of that period. The budget is set according to the formula

The initial amounts are chosen by selecting a random integer in the range [0, 2*(2^Nₚ - 1)] for each asset. Furthermore, we use the parameter values γ=0.5, ν=0.01, ν’=0.001, Nₚ=1, and ρ=0.001*N. For each portfolio optimization instance we choose Nₐ assets randomly from a total predefined selection of 100 assets. Note that as a result of Nₚ=1, the number of qubits for each instance is equal to Nₐ.

For the variational quantum state we use alternating layers of Ry rotation gates (where the angles are the variational parameters) and CNOT gates between subsequent qubit pairs (i.e. linear entanglement). When not using a warm start, the initial parameters are all set to 0 except those of the last rotation layer, which are set to π/2 such that the resulting initial state is |+⟩ⁿ. When using a warm start, we also set all parameters to 0 except those of the last layer, which are in this case defined in terms of the continuous solutions xᵢ,ₛ as

Each quantum circuit is executed 1024 times. For VQE and α-VQE we use COBYLA as the classical optimizer, whereas for f-VQE we implement a standard gradient descent solver with a learning rate equal to 2. In all cases we use a maximum number of 100 iterations. Furthermore, for α-VQE we use α=20% and for f-VQE we use the filtering operator H^-τ with τ=200.

All algorithms are simulated on a classical computer in the absence of a noise model, i.e. we assume a fully fault-tolerant quantum computer.

Quantum(-inspired) improvement of portfolio optimization was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

Summary

• Duplicate claims fraud is estimated to cost insurers billions of dollars each year
• But it is a difficult problem to quantify without a solution that can first accurately detect cases
• ClaimShare Backtesting allows insurers to assess lost value by detecting historical duplicate claims
• Backtesting was conducted that underscored ClaimShare’s accuracy and potential to save national insurers millions in avoidable losses

Duplicate claims cost insurers billions of euros each year, but without detection, it is an invisible problem. Now, however, it is possible to identify the number of potentially fraudulent claims insurers typically receive using ClaimShare Backtesting.

1. A major European insurer signed on for Backtesting with ClaimShare
2. The insurer provided claims data in 5 pre-selected datasets
3. ClaimShare was used to detect duplicate claims

Results of Backtesting

Backtesting took only minutes and detected 100% of the duplicate claims attempts.

Results were in-line with KPMG estimates for the percentage of claims that are likely to be duplicates — roughly 10% of all fraudulent claims.

Not only did ClaimShare process all claims from the 5 pre-selected datasets quickly, but the solution proved perfectly accurate, detecting all duplicates.

This, in turn, made it possible to quantify the financial benefit of detecting the fraud attempts for the insurer. In this case, if used, ClaimShare would have saved the insurer €2.7 million within only the pre-selected data.

As the Backtesting did not include the totality of this insurer’s annual claims, nor of other national insurers, quantifying the significant potential savings for a market-wide consortium using ClaimShare was not possible. It is clear, however, that substantial value would be found through deploying ClaimShare and detecting duplicate claims at this scale.

Companies interested in participating in Backtesting are invited to contact ClaimShare’s Product Manager, Steven Eliaerts via the company website.

To perform Backtesting, the ClaimShare participant simply needs to:

1. Extract the claims dataset for the agreed business line
2. Standardize the PII-Data in the claims dataset to the required data formats
3. Upload the claims dataset to your personal secured ClaimShare environment

Please book a demo for more information on either the ClaimShare solution or Backtesting.

ClaimShare Backtesting: quantify possible double-dipping fraud was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

Summary

• Duplicate claims fraud, or double-dipping, is estimated to cost insurers billions of euros each year
• Insurers are turning to inflexible databases to combat double-dipping and other forms of fraud
• But reliance on centralized infrastructure typically leads to limitations in the scalability, flexibility, and effectiveness of anti-fraud initiatives
• ClaimShare is a novel alternative for data sharing, allowing insurers to uncover hidden value by detecting and preventing duplicate claims with greater effectiveness than ever before

What do an injury in Slovenia, a suspicious series of car accidents in France, and recurring shipwrecks in Corsica all have in common?

The answer: double-dipping fraud.

Double-dipping, or duplicate claims fraud, potentially occurs when a policyholder files the same claim with multiple insurers with the intention of recouping more money than they are legally entitled to. It is a scheme that comes in many different forms.

In 2019, a court found a 22-year-old Slovenian woman guilty of deliberately harming herself to profit through insurance fraud. Having taken out 5 separate insurance policies with different insurers a year prior, she stood to gain more than €1 million in one-off payouts and monthly disability payments.

The woman and a number of relatives were arrested after she arrived in hospital. She and her boyfriend received two- and three-year prison sentences respectively.

In a more typical case in France, a 31-year-old man filed more than 300 accident reports over a four-year period, starting in 2015 and amounting to €400,000 in damages.

The scheme targeted 16 different insurance companies and would generally involve the staging of deliberate car “accidents”. The offender would then offer to pay for repairs, taking the car to an accomplice’s garage which would overcharge for the service, allowing for the maximum payout.

Another example of double-dipping, the “Titanic affair” in Corsica, cost eleven insurance companies even more money — a total of €1.7 million between 2012 and 2017. The racket principally involved a yachtsman, a nautical professional, and a corrupt expert. Together, they collaborated to produce duplicated insurance claims for invented shipwrecks.

In total, prosecutors brought charges against 30 individuals involved in the fraud scheme, after numerous claims were filed using exactly the same details and wording.

These cases are only publicly available because fraudsters were either too lazy or not savvy enough to avoid getting caught. But a huge amount of double-dipping can be difficult to identify. This is a serious problem, as the phenomenon costs the insurance industry a substantial amount of money.

How much?

KPMG estimates that between 5 and 10% of all insurance fraud involves double-dipping. If you have any idea about the scale and cost of fraud for insurance companies globally, then you understand that these percentages are far from trivial, equating to billions of euros in avoidable losses.

The need for collaboration

Double-dipping fraud relies on the fact that it is difficult, particularly for regulatory reasons, for insurance companies to compare details on insurance payouts.

One way insurers have been trying to navigate this problem is through shared databases. The aim is to collect claims data from different insurance companies and store it in encrypted form at a central agency or organization. Claims can then be compared with this bank of information as and when needed.

This approach has certain typical functions and intended benefits:

1. Personal data from claims is encrypted and housed in a central database
2. Personal data for new claims at insurers is encrypted and shared as a link
3. Links are compared against all other claims in the central system, with data falling into different categories (for example, Name, Address, License plate, bank account number)

If the system detects a match, an alert will be sent to the insurer with certain complementary data — the number of other insurers with a matching claim, for example.

One drawback to this approach, however, is that a 100% data match is typically necessary for detection. Another is that it can be challenging to match claims data across business lines — between car insurance and personal injury claims, for example — offering dishonest individuals avenues to avoid detection.

Due to compliance requirements (necessitating the minimization of data) database-based anti-fraud solutions are also not typically rich in data, limiting their functionality and flexibility. Data is also stored centrally: as the past few years have demonstrated, this can present data security vulnerabilities, regardless of encryption techniques.

For the above reasons, databases are not the best solution to combat duplicate claims fraud.

ClaimShare: a unique alternative, ideally suited to data sharing in regulated environments

How is ClaimShare different?

• ClaimShare is not a centralized database. It is an application with layers of technology that overcome the challenges of regulation-compliant data sharing.
• ClaimShare distinguishes between public data and personal identifying information (PII), handling them through completely separate funnels and processes.
• ClaimShare uses blockchain technology and artificial intelligence to match public data in a decentralized and streamlined way. The solution uses confidential computing to match encrypted personal data in a secure Enclave hardware component.
• ClaimShare’s fuzzy matching algorithms also mean that only around 5% of PII needs to be compared, making it inherently more scalable than alternatives.
• ClaimShare does not need 100% matches to detect double-dipping, making it more accurate and harder to fool than alternatives. This also allows the solution to match across business lines, so fraudsters can no longer game the system.

Follow the link to learn more about ClaimShare’s unique technology and value for insurers.

Anti-fraud data sharing: preventing double-dipping requires flexibility was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.

Wednesday 6th April 2022 — Catalyst Blockchain Platform has officially launched its Blockchain Adoption Program for innovation labs, research institutes, and accelerators. From today, organizations at the forefront of innovation can enroll to receive special commercial terms for the use of the platform. Collaborating innovators benefit from a 30% discount and an additional free month on the most complete blockchain management solution available.

Catalyst Blockchain Platform is a comprehensive solution, allowing anyone to build, deploy, and maintain blockchain networks anapplications easily, code-free. Catalyst Blockchain Platform is developed by IntellectEU, a global leader in distributed finance and emerging technologies.

“By collaborating with other organizations, we are pursuing a central goal: spreading blockchain adoption. We believe that blockchain technology has a crucial role to play in the future of business, finance, and application development. That is why we are excited to offer our expertise to more organizations and to guide them in the discovery of what is possible,” said Yana Koldra, Head of Product Management, IntellectEU.

To enroll in the adoption program, visit catalyst.intellecteu.com/innovation-labs

Catalyst abstracts away and automates complex processes for different types of blockchain network configurations, protocols, and infrastructures. The platform is cloud-agnostic, preventing vendor lock-in, and can be deployed on-premises or using blockchain-as-a-service managed infrastructure. The intuitive user interface allows for a smooth management process, even for non-technical users.

Visit catalyst.intellecteu.com to learn more about the platform.

About IntellectEU

IntellectEU is a SWIFT partner and global leader in emerging technologies and digital finance. The company has a reputation for deep expertise in financial messaging and integration, however, in recent years IntellectEU has pushed the frontier of blockchain technology, becoming a founding member of the Linux Foundation’s Hyperledger in 2016. IntellectEU is a Certified Hyperledger Service Provider and the company has a partnership with leading protocol creator R3, retaining an experienced team of R3 Corda-certified developers. For more information about IntellectEU and what the company can do for you, please visit intellecteu.com.

Catalyst Blockchain Platform launches Blockchain Adoption Program was originally published in IntellectEU-blog on Medium, where people are continuing the conversation by highlighting and responding to this story.