RESEARCH DOSSIER
Insurance Data Advantages and Market Power
Abstract. Insurance data have economic value because they improve forecasts of contingent losses, reveal differences in consumer behavior, reduce costs of verification and administration, and support the design of risk-transfer contracts. Information accumulated through claims, renewals, telematics, connected devices, distribution, and interactions can also change the structure of competition. This monograph examines when an insurer's superior data become a productive advantage, when they confer market power, how they influence entry and consumer switching, and how their private value relates to aggregate welfare. It integrates insurance contract economics, the economics of data, empirical evidence on telematics and risk classification, information-based competition, privacy economics, and comparative regulatory analysis. A critical distinction governs the investigation: predictive accuracy, economies of scale, durable competitive barriers, realized economic rents, and consumer welfare are empirically separate outcomes. Data may be nonrival yet remain costly to acquire, clean, label, govern, combine, and validate. Larger records improve estimation only to the extent that observations contribute relevant, independent information under the target distribution; learning advantages can diminish, deteriorate after structural change, or be replicated. Conversely, exclusive histories, network access, cumulative feedback, infrastructure dependencies, and switching frictions may protect an incumbent without a large standalone improvement in predictive accuracy. The synthesis establishes conditional mechanisms linking information to costs, prices, coverage, innovation, exclusion, and regulatory choices. It specifies rigorous identification and welfare-evaluation requirements, distinguishes established evidence from model-dependent predictions, and evaluates international approaches to data portability, connected-product data access, algorithmic governance, and insurance market conduct. The work presents original comparative analysis without claiming a new theory, an original causal estimate, or a proprietary dataset.
Keywords: Insurance economics; insurance data; market power; predictive advantage; proprietary data; information asymmetry; insurance competition; learning economies; telematics; data scale; data portability; switching costs; underwriting; risk classification; personalized pricing; consumer welfare; entry barriers; regulatory governance; privacy; contestability; algorithmic decision-making; insurance technology; comparative regulation.
Introduction: Insurance Data and the Market-Power Question
An insurer operates by assessing uncertain claims, defining contractual protection, collecting premiums, and financing contingent obligations. This process makes information a central input to both risk transfer and competition. At underwriting, data help an insurer estimate the distribution of loss, distinguish covered from excluded events, and choose deductibles, limits, conditions, and prices. At claims settlement, data can identify fraud patterns, verify events, prioritize assistance, and estimate reserves. At renewal, data reveal customer persistence, past premiums, usage, and responses to changing offers. Across all these activities, additional information can improve service while changing the relative bargaining positions of insurers, intermediaries, and policyholders.
The market-power question arises when information is unevenly distributed and cannot be reproduced quickly at comparable cost. An insurer that observes a long claims history or a connected-vehicle feed may predict a particular loss better than a new competitor. The incumbent might offer a lower risk-adjusted price, avoid adverse selection, or provide a more appropriate contract. The same information can support the identification of customers with high willingness to pay, make renewal pricing more personalized, or create a costly dependence on an insurer-specific monitoring system. These mechanisms have different implications. A data advantage is a comparative difference in information or its effective use. Market power is the capacity to sustain prices, margins, contractual terms, or exclusionary practices above a competitive benchmark because alternatives are constrained. Establishing the former does not establish the latter.
Conventional insurance economics demonstrates why information matters to contract formation. Rothschild and Stiglitz (1976), in their analysis of competitive insurance equilibrium, show how private knowledge about risk may generate separating contracts and complicate equilibrium. Subsequent work recognizes that the insurer may also possess informative signals unavailable to the customer. Abrardi, Colombo, and Tedeschi (2025) explicitly analyze competition among differently informed insurers, obtaining pooling and separating possibilities that depend on signal precision, beliefs, and contractual offers. The direction of asymmetric information therefore varies with the information architecture of a specific market.
The broader economics of data is equally instructive. Jones and Tonetti (2020) analyze data as a nonrival input: one observation can often inform multiple decisions without being consumed in its first use. This property creates an economic case for sharing, subject to privacy and other costs. It does not mean all records are interchangeable. A telematics signal from a particular driver, a historically settled liability claim, a hospital's coded diagnosis, and a property catastrophe footprint have distinct origins, quality constraints, and potential uses.
The present monograph examines six linked questions. How does information enter the cost structure of underwriting and insurance administration? When do additional observations improve forecasts and when do returns diminish? Which contractual or technological arrangements permit incumbents to turn informational differences into durable advantages? How can competition and welfare effects be identified empirically? What roles do portability, access, privacy, supervisory safeguards, and fair market conduct play? How should different jurisdictions be compared without confusing a legal right to transfer observed personal data with a general right to an insurer's proprietary predictive model?
The investigation focuses on insurance as an economic institution. It addresses technical modeling only insofar as model performance changes costs and incentives. Statistical validation, insurance technology architecture, and classification fairness each support a broader research program but have separate independent scientific objects. The central conclusions here concern competition, resource allocation, and welfare across international insurance markets.
Research Scope, Method, and Evidentiary Standards
The analysis uses critical integrative synthesis of primary scholarly research, official regulatory instruments, market-study evidence, and insurance-supervisory documents. Three forms of inference are kept distinct. A mathematical proposition is conditional on its assumptions and equilibrium concept. An empirical association is conditional on measured variables and sampling. A causal policy claim requires an identification strategy that isolates the intervention from simultaneous changes in the insurance market. None can be substituted automatically for another.
The core unit of economic analysis is an insurance market defined by the insured exposure, customer population, product design, jurisdiction, distribution channel, and feasible substitutes. "Insurance" as a whole is usually too broad a relevant market for an entry analysis. Automobile insurance telematics, individual health underwriting, catastrophe reinsurance, and small-business liability coverage have different data-generating processes and institutional constraints. A relevant market must also reflect differentiated products and the practical ease of changing contracts and suppliers.
Evidence was selected to test the mechanisms rather than to accumulate favorable examples. It includes classical insurance-information models; empirical studies with divergent selection findings; general data-economics results; specific telematics studies; a recent theoretical analysis of insurer-held signals; the United Kingdom's documented renewal-pricing market study; and European and North American legal or supervisory records. Journal articles and official institutional texts are cited within the relevant claims. The available evidence is heterogeneous in the observation period, population, jurisdiction, outcome, and analytical method, so a pooled effect size would be unjustified.
Formally, denote insurer j's effective information by I_j, its prediction function by f_j, the policyholder's risk characteristics by X, and realized loss by Y. Data advantage can be represented by a difference in expected out-of-sample predictive loss, R_j = E[L(Y,f_j(X))], evaluated on the same target population and at a comparable observation date. A lower R_j indicates better predictions under the chosen scoring rule. The economic implication depends on the loss function, calibration, administrative expenditure, allowed classification variables, legal constraints, and contract response. No empirical value of R_j is estimated in this manuscript. The notation clarifies what a valid comparison would require.
The investigation separates prediction from competition at another level. Let P_j be the premium, C_j the expected cost of claims and administration, and M_j = P_j - C_j a simplified unit margin before capital, acquisition, and other costs. An informational improvement may reduce C_j while market pressure lowers P_j by a similar amount. It may also raise M_j when rivals cannot replicate the improvement, or simply transfer rents between consumers and suppliers when underwriting selection shifts. Observing a more accurate model does not identify the counterfactual trajectory of P_j, C_j, or M_j.
The methodological priority is therefore comparative: evaluate the same economic mechanism under competitive entry, incumbent exclusivity, portability, and differing risk-pooling rules. The original contribution is a structured, claim-specific synthesis of conditions and evidence. This is an interpretive scholarly assessment, not a newly instituted theory or formal model.
Insurance Data as an Economic Asset
What Insurers Observe and What They Can Learn
Insurance information is produced at multiple stages. Applications contain disclosed characteristics and proposed coverage. Underwriting and inspections produce verified descriptions, risk indicators, images, tests, and external reports. Claims generate event descriptions, timing, severity, adjudication records, and adjustments. Renewals record prices offered, acceptance, lapse, complaints, and policy changes. Connected devices can contribute driving patterns, industrial sensor readings, physical activity, or property conditions, depending on the product and jurisdiction. Intermediaries and platforms may observe quotations across competing insurers.
These records should be distinguished according to their economic relationship to the insured risk. A claim event is a relatively direct observation of a realized loss under a particular contract and claims process. A driver's acceleration pattern is a proxy for future accident exposure, whose relevance depends on context and measurement. A customer's response to renewal pricing reveals something about demand but cannot by itself validate expected claims. A claims-handling duration can indicate operational performance while reflecting the complexity of cases selected into the insurer's portfolio.
Even direct records are affected by reporting, thresholds, deductibles, coverage exclusions, claim suppression, administrative coding, and litigation practices. For example, a zero recorded claim may mean that no loss occurred, that the loss was below the deductible, that the person declined to claim, or that a claim has yet to emerge. An insurer with more observations can nevertheless estimate the wrong economic object when the sample includes changing coverage and claims practices. Data quantity becomes economically meaningful only through an explicit observation process.
A useful typology is exposure data, outcomes data, behavioral data, transaction data, and infrastructure data. Exposure data define the denominator of risk frequency. Outcome data identify realized severity and frequency, subject to reporting lags. Behavioral data may proxy both hazard and price sensitivity. Transaction data reveal acquisition and switching. Infrastructure data capture the reliability and cost of collecting and using the preceding categories. Combining the categories can create information that none yields alone. The value of combination depends on legal permission, technical interoperability, and inferential validity.
Data Assets, Property Rights, and Contractual Control
Insurance information is rarely subject to a single undivided property right. Control can arise from private contracts, database protection, intellectual property in software, confidentiality, access credentials, consent, data-protection law, and the physical operation of devices or platforms. An insurer may control a model trained on risk data while a policyholder can exercise rights over relevant personal information. A telematics vendor may hold access to raw sensor streams; an insurer may receive only a score or report. A broker may retain quote histories and customer relationships. Economic control must be specified at the level of each information stream.
It follows that "data ownership" is an incomplete measure of competitive advantage. An exclusive agreement to collect new observations may be more important than possession of a legacy database. The duration of exclusivity, switching costs, ability to infer equivalent information from substitutes, and contractual restrictions on third-party use determine whether the advantage persists. A transfer restriction that preserves security may have a very different consequence from one that blocks contestable entry without demonstrable consumer benefit.
The nonrival character of a datum changes its potential social use. Jones and Tonetti's analysis of nonrival data illuminates the tension between the benefit of wider access and incentives for private generation. A claims record can inform reserving, fraud detection, pricing, and research without physical depletion. Yet duplication of access does not eliminate the expense of secure storage, standardization, identity matching, processing, interpretation, and governance.
Access to the same records does not guarantee equality of performance. Organizations differ in actuarial expertise, computing infrastructure, available complementary variables, distribution channels, and the capacity to respond to model outputs. The empirically relevant asset is an effective information-production capability rather than raw stored volume.
Nonrivalry, Complementarity, and Information Quality
Economic data have several important characteristics: they may be reused, joined with other datasets, and partially replicated through observation of new transactions. They may also contain personal or trade-sensitive information whose use creates obligations and costs. The productivity of additional data is thus conditional on the quality of labeling and the match between training and decision populations. Farboodi and Veldkamp (2026), in A Model of the Data Economy, analyze data creation through economic activity and its consequences for firm behavior. Their work addresses economy-wide mechanisms and is not direct evidence of insurer-specific monopoly power.
Complementarity can magnify informational advantages. A claims database might be much more useful when linked to geospatial exposure, litigation patterns, repair networks, or driving behavior. An entrant that obtains a public loss table may still lack the operational infrastructure needed to integrate predictions with claims and pricing. On the other hand, a standardized external dataset, an industry pool, or a specialist vendor can reduce the amount of proprietary history needed to compete.
The marginal value of data also depends on rare-event structure. A large number of routine small claims may provide limited independent information about infrequent catastrophes. For extreme losses, the relevant data may be concentrated in industry claims aggregations, physical science, hazard models, and reinsurance information rather than the experience of one primary insurer. The relevant bottleneck is often high-quality information about rare scenarios, not the number of consumer records alone.
Another constraint is the feedback between model deployment and observation. Once a new underwriting rule declines certain applicants, the insurer ceases to observe the claims that those rejected applicants would have generated under coverage. A model trained on accepted policyholders may progressively become less representative of the potential market. Data gathered under incumbent selection decisions are therefore not a neutral sample of the customers an entrant might serve.
The Economic Production of Insurance Information
Insurance data are produced jointly with selling policies, collecting premiums, administering claims, maintaining customer relationships, and fulfilling regulatory obligations. The incremental cost of collecting an additional record can be low after the infrastructure is in place, while fixed costs of data architecture and governance remain substantial. This combination supports economies of scale without guaranteeing unlimited increasing returns.
Economic value can be decomposed as the expected operational savings from better forecasts or processes; the expected profit effects of changing underwriting selection and contract terms; the option value of future reuse; and the value of excluding rivals from otherwise feasible access. The first and third may correspond to resource-saving social gains. The second requires an incidence analysis to distinguish savings from transfers. The fourth demands careful evaluation because a firm's exclusionary value can rise when total market welfare falls.
The EIOPA digitalisation survey documents technology adoption and insurer practices across surveyed European undertakings. Its reported usage shares describe respondents in a particular survey, not the prevalence of a specific proprietary-data advantage in all insurers worldwide. In EIOPA's 2024 communication, half of responding non-life insurers and 24 percent of responding life insurers indicated use of AI, a broad category spanning underwriting and operations. This pattern demonstrates adoption heterogeneity, not the size of any competitive moat. Moreover, the reported heavy use of large technology providers for cloud storage places important parts of the information-production chain outside insurers themselves.
Prediction, Information, and Cost Differences
Predictive Improvement and Economic Value
A more accurate insurance forecast is valuable when it changes a costly decision. Better claims-frequency estimates can support rate adequacy, reduce underpricing of concentrated exposures, and allow firms to quote risks that would otherwise be excluded. Better fraud assessment can accelerate legitimate claims and reduce investigation expense. More accurate loss-severity forecasting can improve reserving and liquidity decisions. A risk indicator that moves a probability estimate but leaves underwriting decisions unchanged may have little immediate economic value.
For risk prediction, evaluation requires the proper target: claims under comparable coverage, exposure, inflation, claims handling, and observation windows. In binary classification, a log-loss or proper probabilistic score may measure forecast quality, while profitability depends on premiums, claims severity, operating costs, and constraints. A model can improve a discrimination statistic while remaining poorly calibrated in important subpopulations. A statistically significant lift need not translate into material economic rents.
Let expected insurer cost per exposure be c(x) and the premium p(x). If an improved model changes conditional cost estimates, the insurer may alter p(x), its acceptance rule, deductible, or coverage terms. Under vigorous competition, the long-run equilibrium may pass much of the cost saving to customers, while the innovator initially earns a temporary return. When exclusive information persists and consumers cannot readily compare offers, a larger share of the gains may remain with the incumbent. Price elasticity and switching matter alongside estimation.
Information also creates value by reducing uncertainty around a risk estimate. Two firms may offer equal average expected premiums but hold different confidence about extreme costs. An insurer with stronger data may require less precautionary pricing or capital for a narrowly defined exposure, conditional on solvency rules and true model reliability. This effect must not be conflated with increased risk appetite resulting from overconfidence in a poorly validated model.
Selection, Underwriting, and Contract Equilibrium
The classical adverse-selection problem arises when customers know more about expected risk than insurers and respond to pooled pricing. In Rothschild and Stiglitz's equilibrium analysis, informational asymmetry can lead to differentiated insurance terms, imperfect coverage, or equilibrium difficulties. More precise insurer information may reduce some pricing distortions, but it may also alter risk pooling and the distribution of affordable coverage.
Empirical work disciplines broad claims about selection. Chiappori and Salanié (2000), studying automobile insurance, develop tests for asymmetric information and find little evidence of the predicted conditional positive correlation in their examined French sample. Cohen and Einav (2007), using deductible choices to estimate risk preferences, demonstrate that demand can reflect heterogeneous preferences rather than only private beliefs about risk. The existence and welfare consequences of selection cannot simply be inferred from differences in customer claims histories.
Better information may also change the extensive margin of insurance. If an insurer learns that an applicant has a much higher expected loss than was previously estimated, it may raise the premium or withdraw coverage. This response can improve the accuracy of expected-cost accounting while reducing the applicant's access to risk transfer. Conversely, information identifying safer-than-average individuals can allow previously excluded customers to obtain cheaper coverage. Total welfare depends on the distribution of these changes, willingness and ability to pay, the costs of risk bearing, and alternatives.
The economic review of insurance market models by Einav, Finkelstein, and Levin (2010) emphasizes structural analysis connecting demand, cost, selection, and counterfactual welfare. It is particularly relevant here because testing for a positive correlation between insurance and loss is not itself a measure of the welfare effects of better data. One must recover how participation and contracts would change under competing information regimes.
Informational Advantages on Both Sides of the Contract
The premise that only policyholders possess private information is increasingly inadequate for data-intensive markets. An insurer may have private signals derived from a population-level model, a proprietary device, or a historical archive that a customer cannot interpret. Competitors may obtain different signals about the same exposure. The economic consequences follow from strategic offer design, beliefs, and the option to accept or reject contracts.
Abrardi, Colombo, and Tedeschi (2025) derive information-dependent insurance equilibria in a theoretical setting with imperfectly informed competing insurers. Their equilibria can feature pooling and positive profits; when signals are revealed through offers, underinsurance may arise from the interaction of signaling and screening. Greater precision can exert upward pressure on premiums under specified assumptions. These are model-specific analytical findings, not an empirical estimate of the effects of big data in a named insurance market.
An insurer's informational advantage can also involve demand rather than loss. A provider that tracks renewals can estimate an individual customer's propensity to shop, negotiate, or accept a price increase. This can be combined with actuarial information to set different premiums for customers of similar expected risk. Demand information and risk information should be separated in both economic analysis and supervisory scrutiny: one may improve matching of price to expected cost, while the other may help extract surplus from inertial consumers.
The resulting competitive environment can resemble bilateral asymmetric information. Customers may know effort, habits, future intentions, or health symptoms better than insurers. Insurers may know the statistical implications of disclosed variables, portfolio correlations, or dynamic pricing strategies better than customers. Competitive effects then depend on information about information: the customer's ability to understand offers, independent intermediaries' access to comparison data, and the credibility of third-party estimates.
Operational Efficiency and Pricing Power
The observed unit cost of an insurer combines expected claims, claim settlement expense, underwriting, marketing, technology, capital, and other overhead. A data improvement can reduce losses from misclassification while increasing expense on analytics or device collection. Profitability and welfare must be evaluated net of these costs. A firm with superior predictions may still be inefficient overall if it bears a costly exclusive vendor contract or high distribution expenditure.
To identify market power, one must establish a causal mechanism linking a data asset to the capacity to sustain margins or undesirable terms relative to feasible competition. Relevant indicators include persistence of price-cost margins after entrants adopt comparable tools, bid responses when external datasets become accessible, changes in consumer switching and search, and the competitive effect of interoperability interventions. Higher profits alone are not proof: risk-bearing, capital costs, product quality, and differences in loss experience may account for them.
More data can sometimes intensify competition by reducing uncertainty for smaller entrants. Industry loss data, pooled fraud alerts, open hazard information, standardized consumer claims histories, or cost-effective external models may lower the fixed informational cost of quoting risks. The issue is who controls the relevant data-generation process, at what terms, and whether equivalent alternatives can be accessed with predictable quality.
A meaningful assessment therefore combines three tests: whether information materially improves a decision, whether rivals could achieve a comparable improvement through available alternatives, and whether the resulting advantage survives a competitive response. These are ordinary economic tests of mechanism and persistence, not a claim to a new theoretical classification.
Learning Economies and Data Scale
Sample Size, Information Content, and Model Saturation
The intuition that a larger insurer possesses an advantage because it sees more claims requires qualification. Statistical learning improves with relevant information, measured relative to the problem being solved. In simple settings, parameter uncertainty can decline as effective sample size increases. But observations are often correlated, repeated, censored, or collected under different contractual conditions. Two million near-identical records can contain less useful information about a rare risk than a small number of carefully verified and diverse observations.
Suppose that a model trained on n effectively independent records has expected prediction error E_n = B + V_n + D_n, with B denoting irreducible uncertainty and persistent specification bias, V_n sampling-related error, and D_n mismatch between the training and deployment distribution. Additional relevant records can reduce V_n. They need not change B, and they may increase D_n if the new observations come from a population unlike the intended customers. This schematic decomposition is used to explain standard statistical mechanisms; it is not introduced as an independently estimated model or original theory.
In practice, the shape of the learning curve depends on model complexity, target frequency, label quality, and changing hazard conditions. Learning may be fast initially and then saturate. When a mature actuarial table already captures major risk determinants, proprietary micro-observations may deliver modest marginal accuracy. When a market is new or exposures are highly heterogeneous, access to a small set of otherwise unavailable explanatory variables may matter greatly. What matters competitively is the gap between feasible rivals' performance, not the incumbent's absolute number of data points.
The limits of scale are especially important in insurance because portfolio growth alters the risk population. A large insurer may hold more data but serve a different customer mix, different regions, and different contract forms. Without appropriate adjustment, apparent accuracy improvements can reflect the ease of predicting incumbent-selected customers. A valid test should therefore evaluate both incumbents and entrants on the same ex ante risk population using common outcome definitions.
Dynamic Feedback, Drift, and Heterogeneity
Insurance data form a dynamic loop. A model affects who receives an offer, who buys, how a policy is priced, which claims are presented, and which claims are paid. These decisions determine the next training sample. A feedback loop can reinforce an advantage if a growing insurer observes diverse risks that improve future decisions and attract more customers. It can also narrow the observed sample and conceal systematic errors where rejected, uninsured, or differently serviced customers disappear from records.
Dynamic scale advantages must be evaluated net of model drift. Changes in driving technology, climate exposure, medical practice, inflation, legal liability, repair costs, and fraud behavior can weaken older relationships. A large archive may contain observations whose nominal values, risk categories, and decision context are no longer representative. For rapidly changing exposures, current data and external knowledge can be more informative than historical volume.
Farboodi and Veldkamp's economic treatment of data accumulation provides a general framework for data generated as a by-product of activity. Translating that framework to insurance requires allowing for policy discontinuities, delayed claims, and heterogeneous selection. An incumbent can generate proprietary signals through current operations, while an entrant may purchase third-party services or obtain market information through partnerships. Whether the feedback creates a durable barrier is ultimately an empirical question.
The distinction between firm scale and information scale is essential. Some insurers manage small policy portfolios but access a major reinsurance database or a shared industry loss resource. A large distributor may aggregate quotations across insurers without underwriting the risk itself. A specialized analytics vendor may deliver comparable signals to numerous competitors. These configurations sever any simple mapping from insurer premium volume to a unique informational advantage.
Telematics as an Empirical Test
Automobile telematics provides an unusually concrete setting for evaluating the value of new observations. Data from driving devices or connected vehicles can record patterns such as mileage, time of use, acceleration, and braking. They can supplement traditional variables like age, vehicle type, geography, and past claims. Their economic value depends on predictive performance, device and administration cost, customer participation, and how premiums respond to measured behavior.
Baecke and Bocca (2017) investigate vehicle telematics in risk selection. Their analysis finds that telematics data improve prediction in the studied insurance sample, with economically relevant information emerging during an initial observation period. This is evidence of context-specific predictive value. It does not establish that the resulting information is exclusive, that rival firms cannot obtain comparable data, that policyholders benefit uniformly, or that early adopters earn permanent rents.
Reimers and Shiller (2019) study telematics, competition, and consumer behavior. Their analysis of telematics-related product introduction and competition is especially useful because it considers market responses rather than model accuracy alone. It offers evidence that an innovative data-based product can alter behavior while its early commercial advantage faces competitive pressure. The identification context and geographic market matter; its estimates should not be generalized to health insurance, catastrophe underwriting, or every connected-device market.
The two studies illuminate the difference between technological efficacy and appropriability. A telematics device can generate better forecasts; whether the innovator captures lasting economic rents depends on competitors' access to alternative sensors, their adoption costs, consumer consent, switching, distribution relationships, and regulation. If consumers can choose among multiple telematics programs, better signals may become the competitive standard. If one carrier controls exclusive data through a closed car-manufacturer relationship, the same technology could create entry frictions.
A further issue is self-selection into monitoring. Drivers who opt into telematics may differ from those who decline on unobserved risk, privacy preference, or price sensitivity. Comparing opt-in drivers with other insureds is therefore insufficient to estimate causal risk reduction from monitoring. A program may change driving behavior through feedback, select safer participants, or simply measure risk more accurately. Each mechanism requires its own research design.
Sharing, Synthetic Information, and Data Replication
Data access can arise from transfers, commercial exchanges, industry pools, public hazard information, device operators, and policyholder-directed portability. Sharing may help entrants estimate risks more precisely, improve claims handling, and reduce costly duplication. It may also create privacy, competition, security, and collusion concerns when competitors obtain commercially sensitive information. A database pool should not be assumed automatically procompetitive or anticompetitive.
Access standards can separate data resources from the innovation built upon them. For example, a customer may be able to transmit observed driving records to another insurer while the original firm's proprietary prediction algorithm remains private. The entrant can train its own model or buy analytics, and the incumbent can retain a return to methodological expertise. Whether this is sufficient for effective competition depends on quality, coverage, metadata, historical continuity, and real-time timeliness.
A synthetic dataset or an industry-average model can reduce the need for original records in some settings, but the capability depends on fidelity to rare outcomes, temporal dynamics, and the customer population. A synthetic record containing realistic common patterns may obscure the very tail dependencies or minority-group performance that matter to insurance underwriting. Claims that such data fully eliminate an incumbent advantage require direct validation against the relevant insurance outcome.
Data sharing also changes incentives to collect information. Broad uncompensated access to costly proprietary records may reduce expected returns to investment, while fair, predictable terms can create markets for specialist providers and enable wider social use. Jones and Tonetti's analysis of data-sharing incentives helps explain the economic tradeoff at an abstract level. Insurance-specific institutions must also account for confidentiality, actuarial credibility, solvency, and the costs of maintaining accurate data.
Entry and Market Concentration
Contestability and Exclusive Access
The existence of a concentrated insurance market cannot by itself reveal the contribution of data. Insurers require capital, authorization, reserves, claims-handling capacity, distribution, and reputation. These structural requirements coexist with any informational barrier. The analytical task is to identify what additional constraint is imposed by a proprietary dataset and whether that constraint survives feasible alternative data procurement.
A data-based entry barrier has two elements: access to decision-relevant information materially reduces an entrant's expected unit cost or quote uncertainty, and comparable information cannot be acquired at reasonable cost and speed. A historical claims database that is genuinely unique, properly labeled, and predictive of future losses may meet the first condition. The second may fail if a reinsurer, broker, industry organization, regulator, or data intermediary can supply a functionally equivalent signal.
Exclusivity can appear through device partnerships, distribution-platform interfaces, contract provisions, nonportable records, or access to a closed ecosystem. In a multisided setting, access to policyholders can produce proprietary information that further improves pricing, attracting more policyholders. Yet scale may simultaneously invite specialized entrants targeting a neglected segment with publicly available data and lower administrative costs. A full analysis therefore requires the entrant's feasible strategies rather than assuming an incumbent trajectory.
The Competition Policy for the Digital Era report by Crémer, de Montjoye, and Schweitzer (2019) provides a general framework for investigating data access and competition in digital markets. Its digital-platform analysis is relevant by analogy; it cannot substitute for defining the actual insurance product market or proving exclusion in a particular jurisdiction.
Switching Costs and Data-Dependent Lock-In
An insured customer incurs monetary and nonmonetary costs when changing providers. These include searching for offers, comparing coverage and exclusions, documenting prior losses, reconnecting sensors, adjusting payment arrangements, and learning new claims processes. Some are unavoidable costs of changing contracts, while others arise from how data and infrastructure are organized. Where a new insurer cannot efficiently reconstruct verified history, uncertainty may translate into a less favorable quote or refusal to cover.
Lock-in can persist even when portability is legally available. Data may arrive in inconsistent formats, lack standardized claim definitions, exclude relevant years, or require expensive verification. A customer may have the technical ability to transfer records but lack the attention or confidence to switch. The economic measure of portability is therefore the change in feasible alternative offers and realized switching costs, not merely the existence of an export interface.
Data advantages can also combine with behavioral pricing. An incumbent that knows which customers are unlikely to shop can sustain higher renewal prices for them. Its knowledge of customer inertia then creates a profit opportunity separate from improved risk prediction. The United Kingdom's general insurance pricing market study investigated such price-walking behavior in home and motor insurance, among other problems. The findings demonstrate a market-conduct mechanism, not proof that a specific artificial-intelligence model caused it.
The question of customer surplus is distributional. A highly engaged consumer may benefit from aggressive acquisition discounts, while an inertial customer bears larger renewal costs. Interventions that constrain renewal pricing can change both groups' prices, marketing incentives, and switching patterns. The FCA's later evaluation of pricing-practice remedies evaluates price walking, overall prices, product quality, and switching costs. Each is a distinct outcome; an intervention's effect on one cannot be inferred automatically from its effect on another.
Platforms, Distribution, and Cloud Infrastructure
Information about insured risks often crosses organizational boundaries. Comparison platforms observe quotes and purchase behavior across carriers. Managing general agents and brokers can aggregate underwriting or claims records. Connected-device manufacturers may control sensor interfaces. Cloud and analytics vendors provide common infrastructure to multiple insurers. Market power can therefore be exercised or constrained at layers adjacent to the insurer.
A distribution platform may reduce consumers' search costs by making competing quotes visible, while gaining a privileged view of demand elasticities and conversion rates. It may strengthen competition among insurers yet become an information bottleneck for access to customers. A cloud provider can lower fixed technology costs for small insurers and, through switching costs or limited interoperability, create vendor dependence. The welfare effects of each relationship require evaluation at the appropriate layer.
EIOPA's European insurance digitalisation report found extensive commercial relationships with large technology companies among its survey respondents. This is relevant to systemic dependence and the location of information-processing capabilities, but cloud-service adoption is neither an observed insurance-data foreclosure event nor a measure of insurer-level market power.
Competitive analysis should consider multi-homing, exclusive contracts, interoperability, available intermediaries, technological switching, and the distribution of bargaining power between insurers and vendors. The relevant asset may be a proprietary feed or a customer channel rather than an actuarial model. Policies aimed solely at insurer-held records may leave the dominant competitive constraint untouched.
Identification of Data-Related Market Power
A serious empirical study must establish a counterfactual. One might exploit a staggered rollout of a data source, a regulatory change affecting permitted access, a third-party partnership opening or ending, an unanticipated loss of device connectivity, or a market entry that supplies comparable data capabilities. The outcome should include prices, risk-adjusted costs, coverage, entry, customer switching, and margins, not only model accuracy.
Difference-in-differences designs need defensible common trends absent treatment. If an insurer deploys telematics precisely where claims inflation is accelerating, the observed premium change may reflect market conditions rather than access to information. Instrumental-variable designs need credible exclusion restrictions. An instrument that affects both data availability and driving safety cannot identify only the information channel without additional structure.
A transaction-level approach could compare incumbent and entrant quotes for the same risk under comparable contractual terms, examining whether data access predicts residual pricing differences and whether these differences shrink after interoperable data become available. This design still needs to address unobserved service quality, capital costs, and customers' selection of insurers. Stable claim labels and common evaluation populations are essential.
Market concentration statistics such as firm shares or a Herfindahl-Hirschman index are descriptive context, not direct evidence of informational foreclosure. A concentrated market can have intense potential competition; a fragmented market can include discriminatory pricing and poor switching. A convincing inference would show that a data access change alters the entrants' viable pricing, customer acquisition, or persistence relative to an economically plausible counterfactual.
Abrardi, Colombo, and Tedeschi's heterogeneously informed competition model is a reminder that informational rents can exist even with multiple insurers and that higher concentration need not map monotonically to profits in a particular equilibrium. The conceptual mechanism must be identified before attempting a universal statistical prediction.
A defensible evidence hierarchy begins with a verified change in effective information, continues with a demonstrated decision or cost effect, and proceeds to a separately measured competitive outcome. Inference across those stages requires explicit assumptions, sensitivity tests, and disclosure of missing comparisons. This procedure is a methodological application of standard economic reasoning rather than a claim of independent theoretical invention.
Data Portability and Governance
The Difference Between Transferable Records and Inferred Scores
Data portability changes the economics of entry only if usable information can move to a competing supplier at a cost that permits effective use. Three layers must be distinguished. The first comprises information supplied directly by the customer. The second consists of observed records generated through the customer's use of a connected product or service. The third comprises predictions, risk scores, inferred preferences, and analytical models created from the preceding inputs. Legal access and competitive significance differ across the layers.
Consider a driver insured under a telematics contract. Personal details and recorded driving events can help a new insurer estimate expected loss. The incumbent's proprietary score, however, may reflect a model trained on thousands of other customers and business-specific thresholds. The customer has an economic interest in transmitting relevant observations, while the carrier has interests in protecting model know-how and the privacy of other insured persons. Whether a score must be disclosed, explained, exported, or independently assessed depends on specific legal rights and processing circumstances.
Portability may improve competition without reproducing every model output. If competitors can use observed records to estimate risk and offer comparable products, a proprietary score need not become a universal public asset. Conversely, transferring only a small selection of easily observed facts may have little competitive effect if decisive information remains in closed claims histories or device infrastructures. An effective access regime requires attention to metadata, consistency, identification, retention, interfaces, and delivery time.
The economic balance involves consumer choice, expected investment incentives, privacy externalities, error correction, and the cost of ensuring data accuracy. Acquisti, Taylor, and Wagman (2016), in The Economics of Privacy, explain why private decisions to reveal information can have external consequences and why privacy has economic dimensions beyond individual willingness to pay. These insights apply to insurance markets where one person's behavior and information may affect the classification and prices offered to others.
European Data Protection and Connected-Product Rules
The General Data Protection Regulation provides a right to data portability under Article 20 in defined circumstances: personal data concerning the individual, provided to a controller, processed by automated means, and processing based on consent or contract. The legal conditions and exceptions matter. Portability is not a blanket entitlement to all insurer-held data, every inference, every third-party record, or proprietary analytical software.
The United Kingdom's Information Commissioner's Office explains the distinction in its guidance on data portability: observed raw activity data may count as provided by the individual, whereas profiles created by the controller are generally outside the portability right. Such inferred personal data may nevertheless fall within a separate right of access. The guidance concerns the UK regime and is an interpretive resource rather than a binding interpretation of every European jurisdiction.
Regulation (EU) 2023/2854, the European Data Act, governs specified rights and obligations around data generated by connected products and related services, among other matters. Its general application began on September 12, 2025, with certain provisions subject to distinct dates and transitional arrangements. The act is economically important for connected-vehicle and device ecosystems because a data holder may otherwise control information needed by a potential competitor. It does not automatically open every historical claims database or transfer legal ownership of a proprietary insurance risk model.
Data-protection requirements remain relevant when connected-product information includes personal data. A data-access authorization and permission to process sensitive information are separate questions. Insurance applications can also implicate contractual confidentiality and other sector-specific law. The extent of competitive opening therefore depends on product coverage, user rights, technical access, applicable safeguards, and the business relationships among manufacturers, data holders, policyholders, and insurers.
One can evaluate the effect of these arrangements empirically by comparing the change in quote availability, estimated acquisition costs, switching, and price dispersion in products affected by access rights with comparable products or markets unaffected by the same rights. Such an evaluation needs to account for simultaneous regulatory changes and technology adoption. The legal presence of an access right establishes a possible mechanism; the resulting market-power effect must be measured.
Open Insurance and Interoperability
Open insurance refers broadly to possible arrangements for exchanging insurance-related data and services through interoperable interfaces, often subject to consumer permission and applicable law. EIOPA's open insurance work discusses opportunities and risks from data sharing and APIs. These discussions are useful for designing research questions, but an exploration of open insurance must not be described as a universal, operative mandate that every insurer expose its complete database to rivals.
An effective interface is an institutional agreement, not merely a technical endpoint. Participants must define data fields, authorization, authentication, liability for errors, revocation, timeliness, historical lookback, and permissible downstream use. If different insurers report a loss using incompatible definitions, standardized transport alone may not create useful actuarial comparability. If an interface is costly, unreliable, or difficult for consumers to authorize, nominal portability can coexist with strong incumbent retention.
Data interoperability also has market-structure risks. A dominant intermediary supplying the standard could develop its own bottleneck. A collective pool containing high-resolution future pricing or strategic business information could facilitate coordination if improperly organized. Shared fraud signals can reduce waste but require controls against errors that spread across firms. Evaluation should therefore examine technical functionality, effective competition, and governance simultaneously.
The policy question is which narrowly specified access obligations reduce harmful switching or entry costs while preserving data quality, security, and incentives for investment. Evidence may support standardizing observable personal records in one market and a different approach to specialist underwriting archives in another. The same regulatory instrument cannot be presumed optimal for every line of insurance.
North American Conduct and Algorithmic Oversight
In the United States, insurance regulation is substantially state-based. The National Association of Insurance Commissioners' 2023 Model Bulletin on AI Systems articulates supervisory expectations concerning governance, oversight, and compliance when insurers use AI in consumer-affecting decisions. The NAIC's adoption of a model bulletin is distinct from state adoption and enforceability. Jurisdiction-specific requirements must be determined from the applicable state regulator's actions and laws.
Data-related market power interacts with traditional insurance market conduct because price discrimination, unfair discrimination, underwriting eligibility, claim settlement, and transparency are governed by sector-specific requirements. A large proprietary dataset does not exempt a firm from those requirements. Nor does the existence of an algorithm make every price difference improper. The economically relevant inquiry asks which price differences reflect expected loss, cost, demand elasticity, unfairly restricted competition, or legally prohibited factors.
The U.S. Federal Trade Commission's 2007 study of credit-based insurance scores found predictive association in its automobile insurance data while documenting differences in score distributions across racial and ethnic groups. The report's scope, statistical design, and limitations must be respected: predictive efficacy and distributional effects are simultaneous possible features of a classification system. Its findings are historical evidence about the studied scoring practices, not a current generalization about every AI-based insurer.
North American analysis should therefore keep antitrust, consumer protection, actuarial rate regulation, unfair-discrimination standards, and data privacy distinct. A market may involve several overlapping authorities. International comparisons should not treat a NAIC model text, one U.S. state's implementation, a federal consumer study, and a directly applicable European regulation as legally equivalent instruments.
Welfare and International Regulatory Comparison
Consumer Surplus and Risk-Pooling Effects
The welfare effect of an information advantage depends on the allocation of benefits and burdens among insured individuals, insurers, entrants, and the wider public. A forecast that identifies previously underpriced risks can reduce expected underwriting losses and improve capital allocation; it can simultaneously increase premiums or reduce coverage for riskier households. Better classification may improve prices for safer customers but weaken broad pooling arrangements valued by society for reasons beyond actuarial accuracy.
A conventional expected-utility representation helps clarify this distinction. For customer i, with possible states s, probabilities p_is, premium P_i, gross loss L_is, and indemnity I_is, expected utility can be represented as U_i = sum_s p_is u_i(w_i - P_i - L_is + I_is). Better information changes the feasible premium and indemnity schedule as well as participation. Social evaluation also depends on insurer resource costs, distributional weights, and the viability of risk-bearing arrangements. This is an application of standard contingent-claims and welfare economics, not an original social welfare model.
The effect of more granular risk classes is therefore conditional. If classification reduces cross-subsidies, some customers gain and others lose. For risk-averse customers, a policy that remains available at an actuarially fair but very high premium may offer less practical protection than one supported by credible redistribution or broad pooling. Conversely, a pooled premium may exclude low-risk people who refuse to subsidize high expected claims, raising total participation problems. The suitable baseline depends on the institutional policy objective.
Eling, Gemmo, Guxha, and Schmeiser (2024) examine big data, classification, and privacy. Their research emphasizes that more accurate prediction, the choice of risk classes, and incentives to disclose personal information interact. They show in a modeled setting that higher classification accuracy conditional on private disclosure need not lead insurers to segment the market more finely. This is direct support for evaluating privacy preferences and equilibrium behavior rather than assuming that all predictive improvements imply finer pricing.
Price Discrimination, Loyalty, and Market Conduct
Two insurance customers with equivalent estimated loss can pay different prices because of product characteristics, acquisition channels, contracts, insurer costs, loyalty, and differences in price sensitivity. Identifying the source of the difference is essential before attributing it to data-driven market power. Risk-related classification can serve underwriting accuracy; willingness-to-pay segmentation aims to capture more consumer surplus and may depend on the same analytical infrastructure.
The FCA's General Insurance Pricing Practices report documents problems in UK home and motor renewal markets, including repeated premium increases for some renewing customers. Its monetary estimates are specific to the underlying observation period and counterfactual; they should not be reported as a timeless measure of global insurance data rents. The subsequent remedies evaluation illustrates why outcomes should be tracked separately: stopping tenure-based price walking need not imply that every consumer receives a lower price or that product quality and switching respond identically.
Market power can therefore manifest as opacity and inertia even with multiple nominal suppliers. A customer's ability to obtain a comparable quote is a stronger test of contestability than the theoretical existence of alternative insurers. Intermediaries, comparison tools, cancellation procedures, and renewal disclosures can materially affect the transmission of cost-saving innovation to consumers.
A robust welfare inquiry should report changes in total expenditure, price dispersion among equal-risk consumers, changes in cover and deductibles, claim quality, renewal persistence, and participation among vulnerable groups. Average premium alone can hide reallocation from one customer category to another. Reinsurance and reserve costs must also be considered when pricing practices alter the insurer's risk mix.
Distributional and Privacy Effects
Risk classification can redistribute the costs of uncertainty across households and firms. A new observable indicator may lower premiums for one segment and raise them for another, with the aggregate weighted impact depending on risk aversion, income, and policy objectives. A distributionally neutral prediction model is not necessarily neutral in its outcomes because insurance costs and coverage are inherently state-dependent. Evaluation should disclose whose welfare weights are used and how uninsured or newly excluded applicants are treated.
Privacy is also an economic resource. Individuals can incur costs from collection, retention, sharing, and future reuse of data, even when a monitoring program offers an immediate discount. The economic literature on privacy emphasizes uncertainty about downstream uses and informational externalities. An insurer's private calculation of profit from a new data stream may omit the consumer's cost of monitoring or the impact on people whose classifications become more predictable through others' disclosures.
A data-sharing policy can change both competitive pressure and privacy risk. Portability under user control can improve choice, but many insurers and vendors receiving the same records may increase exposure to data breaches or unexpected secondary use. An access mandate should thus be paired with purpose limitations, authentication, retention discipline, supervision, and enforceable remedies where required by law.
The ethical and legal review of algorithmic systems intersects with competition but should not be collapsed into it. A model can be accurate and competitively beneficial yet violate privacy or discrimination requirements. A model can be fair under a specified evaluation criterion while helping a firm exploit customers unable to compare products. The relevant public-interest analysis considers each dimension explicitly.
Comparative Institutions and Regulatory Tradeoffs
European insurance data governance combines horizontal rules governing personal data, connected products, and AI with sector-specific supervisory and insurance obligations. The GDPR regulates personal-data processing and defined portability rights; the Data Act addresses certain connected-product and data-access relationships. Their scopes differ, and neither instrument alone determines the competitive structure of every insurance market.
The EU Artificial Intelligence Act specifically identifies certain AI systems used for risk assessment and pricing in relation to life and health insurance for natural persons as high-risk in Annex III. The European Commission's AI Act implementation guidance identifies the relevant phased dates, including December 2, 2027 for the specified high-risk Annex III rules as described in its November 2026 guidance. AI Act classification concerns intended uses and legal requirements, not a finding that a particular insurer possesses market power. Regulatory timing must be checked against the law in force and official guidance at the point of application.
EIOPA's AI governance opinion articulates supervisory perspectives on risk management and governance beyond the most salient high-risk categories. Its status as an opinion must be distinguished from binding law. The 2019 thematic review of big data provides earlier sectoral evidence on insurer use cases and consumer protection concerns. A thematic review provides supervisory information, not proof that particular datasets foreclosed competition.
In the United States, the NAIC AI Model Bulletin provides a model for state-level governance expectations and supervisory examination. Separately, competition law, insurer rate regulation, and data-specific requirements depend on jurisdiction. The UK's FCA approach offers a further distinct example of direct intervention against demonstrated renewal-pricing harm, monitored through an ex post policy evaluation.
The comparative conclusion is institutional rather than jurisdictional ranking. A portability right may address customer switching; an AI governance obligation may address model reliability or discrimination; a pricing intervention may address renewal exploitation; competition enforcement may address exclusionary access arrangements. Their purposes overlap but cannot be reduced to a single measure of data advantage. Choice of intervention follows the identified market failure and the costs of remedying it.
Original Analytical Synthesis
The sources support five integrated analytical conclusions. First, information quality and information exclusivity are separate determinants of economic advantage. Better forecasts can create productive efficiency, while exclusive control determines how much of that efficiency a firm can appropriate. A database may be exclusive yet yield little useful risk improvement; a highly predictive external signal may be available to every insurer. Both cases defeat the assumption that data volume translates directly into a durable moat.
Second, data scale generates an advantage only relative to the entrant's feasible information set and the current target population. Statistical saturation, access to third-party information, standardized claims or exposure definitions, and rapid model drift can reduce an incumbent's lead. Proprietary distribution or sensor contracts may preserve access asymmetry even when the underlying prediction technology is replicable. The competitive mechanism should be specified at the level of the actual data stream and contract.
Third, insurer-held information can influence the nature of risk pooling and contract offers, not merely pricing precision. The classical adverse-selection literature addresses customer private information; Abrardi, Colombo, and Tedeschi (2025) show that insurer-side informational differences have distinct equilibrium implications. This matters for policy because compulsory disclosure, greater prediction precision, and more competitors need not have effects that are monotonic in every equilibrium model.
Fourth, evidence of persistent market power requires a measurable competitive counterfactual. The telematics literature provides both predictive improvements and evidence of entry eroding early profit gains. Baecke and Bocca's telematics risk-selection evidence concerns prediction, whereas Reimers and Shiller's competitive response evidence concerns market conduct, profits, and behavior. Together they support a research design that observes the full chain from data to decisions to market outcomes, rather than identifying market power from adoption or prediction alone.
Fifth, the welfare analysis must keep individual insurance protection, product availability, insurer cost, consumer surplus, market entry, privacy, and distribution separate before integrating them. More information can improve allocation and expand coverage for some risks, raise prices for others, and create new monitoring costs. The relevant outcome depends on regulatory and contractual institutions, not on a universal presumption for or against finer risk classification.
An appropriate empirical study would pre-register the relevant insured population; distinguish observed variables from inferred scores; evaluate model improvements on a common time-stamped holdout; account for effective coverage and claims reporting; measure rival data alternatives and acquisition costs; identify exogenous variation in data access or competitive entry; estimate risk-adjusted prices, costs, margins, switching and participation; and include a distributional and privacy assessment. This sequence follows existing scientific methods. It is an applied framework for evidence evaluation, not a newly named theory, doctrine, or model.
Research Limitations and Verification
The evidence establishes mechanisms and conditional findings, with important limits. Several foundational articles were evaluated through journal records, abstracts, or accessible publisher text, rather than complete replication of the original data and code. The telematics studies involve particular markets, observation periods, products, and identification assumptions. The recent heterogeneous-information equilibrium analysis is theoretical. Supervisory surveys describe respondent practices and cannot identify the causal effect of access to proprietary data on insurer margins.
The manuscript does not contain an original microdata analysis, systematic meta-analysis, legal advice for a specific case, new risk-scoring algorithm, or universal estimate of insurance market concentration. Numerical survey shares and policy dates are presented only with source-specific scope. Legal instruments and soft-law opinions are distinguished, as are jurisdiction-wide rules and model texts requiring further adoption. The assessment of the EU AI Act's phased application reflects the cited official materials available on November 14, 2026, and requires no claim about later implementation decisions.
The analysis also identifies areas for further primary research: causal effects of portable telematics histories on consumer quotations; entrant performance before and after access to verified claims data; the role of manufacturer exclusivity and cloud switching; post-intervention renewal price and quality distributions; and insurance-specific links between data-sharing rules, innovation, and privacy outcomes. These are research questions, not outstanding defects in the present synthesis.
Conclusion
Insurance data advantages emerge from the ability to observe, interpret, and use information that improves decisions about uncertain losses and contracts. Their economic significance depends on the relevance of observations, the credibility of outcomes, analytic capability, access arrangements, and the institutional setting of insurance. The same improvement in prediction can lower insurer costs, broaden insurability, change the allocation of premiums, improve claims service, or support greater rent extraction. These effects can coexist and must be measured separately.
A durable data-based market-power claim requires evidence that the informational advantage cannot be replicated or bypassed at comparable cost, that it changes firms' feasible competitive responses, and that the resulting economic rents or restrictions survive entry and switching. Data concentration, insurer size, adoption of AI, and predictive accuracy are individually incomplete indicators. Evidence from telematics demonstrates that accuracy gains can matter while early profit increases may be eroded by competitors. Models of insurer-held private signals further show why changes in information precision can affect equilibrium contracts and premiums in complex ways.
Portability, connected-product access, standardized interfaces, model governance, sectoral rate controls, and competition law address different parts of this problem. Their welfare effects depend on implementation and must account for privacy, risk pooling, consumer choice, and incentives to invest in information. The strongest general conclusion is a requirement of precise causal attribution: identify the valuable information, establish whether and why access is restricted, measure the resulting competitive response, and evaluate who gains and loses under a credible counterfactual.
Research on insurance data and market power consequently belongs at the intersection of insurance economics, industrial organization, actuarial evidence, and comparative regulation. Its scientific contribution is to evaluate mechanisms and their observed boundaries with explicit assumptions and verified sources. The present dossier supplies that integrated assessment while preserving the distinction between established results, conditional theoretical implications, and questions for future empirical testing.
References
Abrardi, Laura, Luca Colombo, and Piero Tedeschi. 2025. "Pretending Ignorance Is Bliss: Competing Insurers with Heterogeneous Informational Advantages." The Review of Financial Studies 38 (7): 2005-2033. https://doi.org/10.1093/rfs/hhae079.
Acquisti, Alessandro, Curtis Taylor, and Liad Wagman. 2016. "The Economics of Privacy." Journal of Economic Literature 54 (2): 442-492. https://doi.org/10.1257/jel.54.2.442.
Baecke, Philippe, and Lorenzo Bocca. 2017. "The Value of Vehicle Telematics Data in Insurance Risk Selection Processes." Decision Support Systems 98: 69-79. https://doi.org/10.1016/j.dss.2017.04.009.
Chiappori, Pierre-Andre, and Bernard Salanie. 2000. "Testing for Asymmetric Information in Insurance Markets." Journal of Political Economy 108 (1): 56-78. https://doi.org/10.1086/262111.
Cohen, Alma, and Liran Einav. 2007. "Estimating Risk Preferences from Deductible Choice." American Economic Review 97 (3): 745-788. https://doi.org/10.1257/aer.97.3.745.
Cremer, Jacques, Yves-Alexandre de Montjoye, and Heike Schweitzer. 2019. Competition Policy for the Digital Era. Brussels: European Commission. Official report.
Einav, Liran, Amy Finkelstein, and Jonathan Levin. 2010. "Beyond Testing: Empirical Models of Insurance Markets." Annual Review of Economics 2: 311-336. https://doi.org/10.1146/annurev.economics.050708.143254.
Eling, Martin, Irina Gemmo, Danjela Guxha, and Hato Schmeiser. 2024. "Big Data, Risk Classification, and Privacy in Insurance Markets." The Geneva Risk and Insurance Review 49 (1): 75-126. https://doi.org/10.1057/s10713-024-00098-5.
European Commission. 2026. "Navigating the AI Act." Official guidance as accessed November 14, 2026. Commission AI Act FAQ.
European Insurance and Occupational Pensions Authority (EIOPA). 2019. Big Data Analytics in Motor and Health Insurance: A Thematic Review. EIOPA thematic review.
European Insurance and Occupational Pensions Authority (EIOPA). 2024. Report on the Digitalisation of the European Insurance Sector. April 30. EIOPA digitalisation report.
European Insurance and Occupational Pensions Authority (EIOPA). 2025. Opinion on Artificial Intelligence Governance and Risk Management. EIOPA opinion.
European Insurance and Occupational Pensions Authority (EIOPA). n.d. "Open Insurance." Official thematic portal, accessed November 14, 2026. EIOPA open insurance.
European Parliament and Council. 2016. Regulation (EU) 2016/679, General Data Protection Regulation. Official Journal of the European Union. Official GDPR text.
European Parliament and Council. 2023. Regulation (EU) 2023/2854 on Harmonised Rules on Fair Access to and Use of Data (Data Act). Official Journal of the European Union. Official Data Act.
European Parliament and Council. 2024. Regulation (EU) 2024/1689 Laying Down Harmonised Rules on Artificial Intelligence (AI Act). Official Journal of the European Union. Official AI Act.
Farboodi, Maryam, and Laura Veldkamp. 2026. "A Model of the Data Economy." The Review of Economic Studies, advance publication, rdag074. https://doi.org/10.1093/restud/rdag074.
Financial Conduct Authority (FCA). 2020. General Insurance Pricing Practices: Final Report, MS18/1.3. London: FCA. FCA final market-study report.
Financial Conduct Authority (FCA). 2025. Evaluation Paper 25/2: Our General Insurance Pricing Practices Remedies. Published July 22, updated December 3. London: FCA. FCA remedies evaluation.
Federal Trade Commission (FTC). 2007. Credit-Based Insurance Scores: Impacts on Consumers of Automobile Insurance. A Report to Congress. Washington, DC: FTC. FTC report to Congress.
Information Commissioner's Office (ICO). n.d. "Right to Data Portability." UK GDPR guidance, accessed November 14, 2026. ICO portability guidance.
Jones, Charles I., and Christopher Tonetti. 2020. "Nonrivalry and the Economics of Data." American Economic Review 110 (9): 2819-2858. https://doi.org/10.1257/aer.20191330.
National Association of Insurance Commissioners (NAIC). 2023. Model Bulletin: Use of Artificial Intelligence Systems by Insurers. Adopted December 4. NAIC model bulletin.
Reimers, Imke, and Benjamin R. Shiller. 2019. "The Impacts of Telematics on Competition and Consumer Behavior in Insurance." The Journal of Law and Economics 62 (4): 613-632. https://doi.org/10.1086/705119.
Rothschild, Michael, and Joseph Stiglitz. 1976. "Equilibrium in Competitive Insurance Markets: An Essay on the Economics of Imperfect Information." Quarterly Journal of Economics 90 (4): 629-649. https://doi.org/10.2307/1885326.