Which insurance options would be considered a risk sharing arrangement?

5.1 Financial sector development

4.3.1 Proportional Risk Sharing

The simplest form of risk sharing is proportional risk sharing, where a plan is paid a fixed combination of a prospective component and a cost-based payment. Suppose the population average spending is x¯and the spending during a year for person i is xi. In a proportional risk sharing arrangement, the plan is paid λx¯+(1−λ)xifor person i where (1–λ) is the portion of the risk (cost) retained by the regulator. As λ approaches 1 the payment system becomes fully prospective, and as λ approaches 0 it becomes fully cost-based. The incentive properties of what has been referred to as a “mixed system” were first studied in the context of provider payment (Ellis and McGuire, 1986),7 and were generalized and applied to plan payment by Newhouse (1996). The prospective component, x¯from above, can be risk-adjusted. Box 4.3 describes a modality of proportional risk sharing currently applied in Belgium.

Box 4.3

Proportional risk sharing in Belgium

In Belgium the total budget for healthcare expenditures (called ω here) is distributed over the insurers according to a weighted formula:

Fv=(RASvω)rω+Ev∑kEk(1−r)ω,

The group of members for whom some risk is shared: the entire population

The types of care for which the risk is shared: all services covered

The extent of the risk that is shared: 1–r

The price that insurers have to pay to share some risk: (RASvω)(1−r)ω

See Chapter 7, Risk Adjustment in Belgium: Why and How to Introduce Socioeconomic Variables in Health Plan Payment, for more details about health plan payment in Belgium.

Some analytic properties of a mixed system can be described easily, useful for introducing the power of risk-sharing methods for improving the performance of a health plan payment system. The key observation is this: a little bit of proportional risk sharing can substantially improve the fit between payment and cost. To make this concrete, consider the risk-adjustment model used for prospective payments to Medicare Advantage plans in the United States (described in Chapter 19: Medicare Advantage: Regulated Competition in the Shadow of a Public Option) containing about 100 variables and decades in development. The R-squared statistic describing the fit of predicted to actual spending at the person level is about 0.11. Proportional risk sharing of only 6% in combination with a simple flat prospective payment set equal to the population average, with no risk adjustment at all, can attain the same fit at the person level. Specifically, a mixed system in which a health plan is paid 94% of the overall population average and then 6% of the actual costs of an enrollee produces a fit of payments to costs with an R-squared equivalent of 0.11.8 This level of fit can be calculated and holds true for any underlying data. In other words, the R-squared equivalent of a proportional risk-sharing plan can be calculated without any data analysis (see Box 4.4).

Box 4.4

The R-squared equivalent of a mixed system

R-squared measures the portion of variation in costs “explained” by the risk-adjustment model predictions. In a mixed system, the “prediction” is a weighted average of the population mean and of actual costs. Writing the mixed system in general form with a weight of λ on the population mean cost and (1–λ) on the individual’s realized cost, the predicted value for person i is: x˜i=λx¯−(1−λ)xi. The R-squared equivalent of explained variance can be expressed as usual as 1 minus the “unexplained variance” as a function of λ:

R-squared of mixed system with (λ)=1−∑i(xi−λx¯−(1−λ)xi)2∑i(xi−x¯)2=1−λ2.

In the text we mention the example of λ = 0.9 (90% of the weight on the mean) and the formula shows R-squared equivalence is 1–(0.9)2 or 0.19.

A mix with more proportional cost sharing can easily outperform the current Medicare Advantage model in terms of person-level fit. A mixed system 90/10 with 90% of the weight on the population average and 10% on costs produces an R-squared equivalent of 0.19, almost twice the fit of the current model. To give just one more example, a 50/50 mix produces an R-squared equivalent of 0.75, vastly exceeding the fit of any risk adjustment formula. Box 4.4 shows the general formula relating the degree of mix to the R-squared equivalent.

While R-squared is an appropriate measure for indicating the extent to which a certain payment system mitigates a plan’s financial risk, it may not be the best metric for measuring the extent to which a payment system mitigates selection problems. As will be discussed in Chapter 5, Evaluating the Performance of Health Plan Payment Systems, the appropriate selection metric depends on the particular selection problem to be analyzed and the underlying market mechanisms. For example, when it comes to measurement of incentives for health plans to select in favor or against particular groups, predictive ratios and measures of under/overcompensation are more meaningful than the R-squared, improvements in these measures can also be readily described when risk-sharing takes the form of a simple mix.9 Our purpose beginning with proportional risk sharing is to preview the power of risk sharing generally, and to set the stage for the value of risk sharing targeted at the high-cost range of the cost distribution.

6.4 Credit constraint and risk management

Credit markets play an important role in agricultural risk management. Here, we focus on “net borrowers” as savings are a type of investment asset modeled in Section 5.3.48 Specifying χt = −Bt where Bt ≥ 0 is the amount borrowed and h (⋅) = κt+1(⋅) ≥ 0 as the repayment function, we have

(6)V(Wt)=maxxt,Bt≥0U(mt)+EtVt+1(Wt+1)

subject to

Wt=xt−Bt+mt,Wt+1=pt+1F(xt,ϵt+1)−κt+1(Bt,K;rt+1,pt+1,ϵt+1),

and

Bt≤(K).

Here K is the amount of collateral, rt+1 is the interest rate (which may be a function of the collateral), and B-(K)is the credit limit faced by the agent. For simplicity, we assume that the level of collateral K is exogenous. The repayment function is given by

κt+1(Bt,K;rt+1,pt+1,ϵt+1)=Kif one defaults(1+rt+1)Btotherwise

which will be driven by price and production risks since they can influence the default decisions.

It is interesting to study credit decisions as a standalone problem, but we also extend the model to consider interactions with insurance decisions in which the model becomes

(7)V(Wt)=maxxt,Bt,θt≥0U(mt)+EtVt+1(Wt+1)

subject to

Wt=xt−Bt+πt(θt,xt)+mt,

Wt+1=pt+1F(xt,ϵt+1)−κt+1(Bt,θt,K;rt+1,pt+1,ϵt+1)+ξt(θt,xt;pt+1,ϵt+1),

and

Bt≤(K,θt+1)

where the interest rate, rt+1, and the credit limit, B-, can now be endogenously influenced through the producers insurance choice, θ. Ultimately the extent of this endogeneity will be context dependent as determined by available loan contract design.

Without deriving the conditions that characterize the optimal choices of xt, Bt, and θt, we can draw useful insights from Model (7). First, from Eq. (3) of Model (2), we can infer that θt is more likely to be zero when Wt is small as st+1 becomes smaller. In other words, the liquidity constraint can play an important role in the demand for insurance. A recent study by Casaburi and Willis (2018) show that the liquidity constraint can deter the demand for insurance. Therefore, access to credit can affect the demand for insurance but it will apparently depend on the structure of offered credit contracts.

On the other hand, availability of insurance can affect credit accessibility particularly through the alleviation of “risk-rationing” (Boucher, Carter, & Guirkinger, 2008). Relating to the endogeneity of loan terms via insurance decisions mentioned above, offering insurance contracts has been considered as a tool to improve credit accessibility (e.g., Miranda & Gonzalez-Vega, 2011). Carter et al. (2016) provide a theoretical foundation for the effectiveness of standalone insurance and credit-interlinked insurance under different agricultural and financial environments. Giné and Yang (2009) present an experimental result that shows that linking a credit contract with insurance reduces the uptake of the credit contract, a finding consistent with Carter et al. (2016) which emphasize the role of agronomic and financial factors. The interaction between insurance and credit can also be particularly important for smallholders and poor households because the presence of uninsured risks and the lack of credit access can be associated with “poverty traps” (e.g., Carter & Barrett, 2006; Dercon, 1998; Lybbert, Barrett, Desta, & Layne Coppock, 2004).49 Mishra et al. (2021) find positive effects of both standalone and credit-interlinked insurance on the likelihood of loan approval in Ghana with a larger impact associated with interlinked insurance. However, offering credit-interlinked insurance contracts can face logistical challenges (e.g., Ahmed, McIntosh, & Sarris, 2020).

Empirical challenges for drawing causal linkages from insurance to credit (or vice versa) exist stemming from the simultaneous nature of the decisions. Motivated by risk-balancing theory (e.g., Featherstone, Moss, Baker, & Preckel, 1988), Ifft, Kuethe, and Morehart (2015) provide empirical evidence on the positive association between the US Federal Crop Insurance Program participation and short-term farm debt which is consistent with the risk-balancing theory and acknowledge the challenge of establishing causality.50 In the context of a tobacco insurance program in China Cai (2016) shows the positive impact of insurance provision on borrowing by leveraging the natural experimental aspect of the program.

10. RELATED LITERATURE

The central advantage of production economies for the understanding of the pattern of financial returns is the added discipline they present to the exercise. Since the actions of the same economic agents give rise to both macroeconomic and financial phenomena, it is a minimum expression of consistency that the same model be expected to replicate the financial and macroeconomic stylized facts, at least along a limited set of dimensions. This has been our perspective. In this section, we discuss other theoretical contributions with significant labor market features and their implications for financial return data. Related comments focusing on empirical contributions to the literature may be found in the essay by John Cochrane in this volume.

Matching financial data in a production setting requires that the capital owner display a strong desire to smooth his consumption intertemporally (provoked by, e.g., a habit formation feature) while simultaneously acting in a context that makes it difficult to reallocate labor or capital to that same end. These latter restrictions essentially substitute for some form of market incompleteness: in either case, agents are prevented from smoothing their consumption across states and dates. In most models, it is the degree of restrictiveness in the labor market that ultimately holds sway vis-a-vis financial characteristics. There are four models, in particular, that we review; principal comparative output data is provided as available. In all cases, notation is harmonized to be consistent with that adopted in this paper.

A first paper to emphasize the influence of labor market phenomena on equilibrium financial returns was Danthine et al. (1992). It proposed a model with shareholders, primary and secondary workers. These latter groups hold no securities (limited participation incompleteness). The primary workers are assumed to have a permanent, full employment association with the firm. Their compensation is governed by a risk sharing arrangement identical to the one proposed in this paper. At the other extreme, the secondary workers' employment prospects are governed by a pure Walrasian mechanism, one that otherwise would lead to substantial income variation. In order to moderate this wage income variability, primary worker wages are postulated to be subject to a wage floor augmented by unemployment compensation (the wage floor is above the market clearing wage in some states) financed by a tax on corporate profits.22 As a result of these latter arrangements, all workers in the model experience income volatility less than what would occur under a full Walrasian scenario. Whether directly—via wage insurance—or indirectly—via the unemployment tax—the net effect of worker income stabilization is to shift income risk onto the shareholders. The principle model results are presented in Table 12.

TABLE 12. Model Results: Danthine et al. (1992)i

Financial and aggregate statistics(a)ii(b)re4.560.84rf3.980.80rp0.580.06ddiii—5.36(d)corr(re, Ct+1/Ct)0.06(e)(f)Output1.760.69Total consumption0.340.98Shareholder consumption5.360.99Investment6.080.99Wagesiv0.220.10Capital stock0.540.03

While the model is able to replicate the stylized business cycle facts very well and produces a premium substantially in excess of what is obtainable under a Hansen (1985) construct, the premium obviously falls significantly short of what is observed. Security return volatilities are also much too low. In effect, variable equilibrium labor supply in the secondary sector in conjunction with shareholder control over investment together provide too much opportunity for shareholder consumption smoothing. Indeed, shareholder consumption volatility is about half the level of the benchmark case of this paper (Table 3, Case (1)); otherwise, the macro series are very similar. In a sense, the current model is a simplified version of Danthine et al. (1992) where all workers are subject to the primary worker income determination mechanism, augmented with an extra source of risk affecting the mechanism of income sharing itself. This second source of uncertainty is fundamental to its superior results along the financial dimensions.23

Boldrin and Horvath (1995) propose a contracting mechanism that is similar to Danthine et al. (1992). In equilibrium, it also has the consequence that employees supply resources to firm owners in high-income states and receive payments from them in low-output ones.24 In their setup, profits and hours both display high levels of variability in line with their respective empirical counterparts. As they do not present data on the pattern of financial returns characteristic of their model, it is difficult to directly compare their results with the other literature. By the nature of their model formulation, however, it is likely that their results would be similar to those in Danthine et al. (1992).

Subsequent to Danthine et al. (1992), the literature approached the same set of issues more from the perspective of modifying shareholder preferences in order that they act in a more risk-averse fashion and less from the “operating leverage” perspective of worker income insurance. The first paper in this tradition was Jermann (1998), which postulates a representative agent style model with habit formation (leading to high MRS volatility) in conjunction with capital adjustment costs that make it difficult to smooth consumption via investment variation. The inability of the agent to smooth is strengthened by a fixed labor supply assumption. With these features, his model is able to explain the business cycle stylized facts in conjunction with the mean premium quite well, but at a cost of excessive risk-free rate volatility. See Table 13.

TABLE 13. Model Results: Jermann (1998)i

Financial and aggregate statistics(a)ii(b)re7.0019.86rf0.8211.64rp6.18—dt+1/dt—8.44(e)Output1.76Consumption0.86Investment4.64

Boldrin et al. (2001) demonstrate, however, that the high premium in Jermann (1998) is significantly reduced if a Hansen (1985)-style labor-leisure choice mechanism is introduced even while retaining the same adjustment cost specification. Thus modified, Jermann's (1998) model also has the counterfactual feature that hours and output are negatively correlated. In this modified model, there are two opportunities for the representative agent (and therefore the representative shareholder) to smooth his consumption stream—by adjusting his hours and investment (though at a cost)—and, taken together, these are very effective consumption smoothing devices. As a result, the premium declines to 0.30 percent. The results in Jermann (1998) are thus not extremely robust. Jermann (1998) is nevertheless, important for establishing the basic modeling perspective for finance cum production models: make the security holders extremely desirous of smoothing their consumption in tandem with technological impediments to doing so.

Boldrin et al. (2001) also review a number of possible model features and ultimately explore one with two sectors—one producing consumption and the other capital goods—where the allocation of capital and labor to each sector must be chosen one period in advance of knowing the respective technology shocks. This has the consequence of reducing the ability of shifts in either factor of production to be used to smooth consumption significantly. It is the restrictions on labor market flows between sectors subsequent to shock realizations, in particular, that they view as most crucial to their results. In conjunction with standard habit formation preferences, these authors can explain the mean equity premium although investment volatility is a bit too low and the risk-free rate again displays excessive volatility, so much so that its standard deviation substantially exceeds that of the return on equity (see Table 14). We note that the excessive risk-free rate volatility of Jermann (1998) and Boldrin et al. (2001) is not a general consequence of the distributional risk perspective.

TABLE 14. Model Results: Boldrin et al. (2001)i

Financial and aggregate statistics(a)ii(b)re7.8318.4rf1.2024.6rp6.63—(e)(f)Output1.97—Total consumption1.360.76Investment4.710.96Hours1.580.78

Danthine and Donaldson (2002) revisit the original question posed in Danthine et al. (1992): to what extent can operating leverage cum income share variation simultaneously explain the business cycle and financial market stylized facts? It is an exploration that is accomplished in a slightly more abstract setting than in Danthine et al. (1992), whereby the latter's elaborate labor market setup (temporary and permanent classes of workers, etc.) is summarized by a “net” risk sharing mechanism nearly identical to the one considered here. The present paper may be viewed as decomposing the general results in Danthine and Donaldson (2002) into the distributional and aggregate shock-related components. With several additional features, such as costs of adjusting the capital stock, Danthine and Donaldson (2002) also achieve an excellent and broad-based fit to the data (Table 15). Generally speaking, Table 15 replicates the one presented in Table 3 except that returns seem to conform to the data slightly less satisfactorily. This is attributable to the slightly lower W/Y ratio, which results in a more modest operating leverage effect.

TABLE 15. Model Results: Danthine and Donaldson (2002)i

Financial and aggregate statistics(a)ii(b)(c)re5.9222.200.26rf2.464.050.02rp3.4622.34—dt+1/dt—16.72—(d)W/Y0.694.83−0.022(e)(f)Output1.77Total consumption1.450.96Shareholder consumption11.940.38Investment3.050.93Capital stock0.27−0.005

Our final theoretical comments concern Guvenen (2005). He assumes a perspective that may be viewed as providing an alternative macro interpretation for the variable risk sharing feature of the present model. Rather than assuming workers and shareholders interacting in an uncertain bargaining context, Guvenen (2005) presumes that the population is divided into two groups with unequal financial market access.25Shareholders participate in both stock and bond markets while non-shareholders trade only bonds. Both groups supply labor inelastically to the firm, and non-stockholders are modeled as being more risk-averse.26 With bond trading being their only mechanism for consumption smoothing, non-stockholders bid up bond prices, resulting in a low risk-free rate. In equilibrium, stockholders end up insuring non-stockholders by increasing their debt holding exactly when a low productivity realization reduces both agents' income, and vice versa. As a result, bond market events act to create a high level of volatility of shareholder consumption, volatility against which they can insure only via management of the capital stock. Although the effective extent of income insurance provided by shareholders to non-shareholders is not as great as in the present model, the fundamental idea is the same. Guvenen (2005) also goes on to show that the consumption of non-shareholders serves a role similar to that of a slow-moving habit in his equilibrium asset pricing equation, a feature also present in our distributional risk sharing formulation. We note that these results seem to be more favorable vis-a-vis “distribution risk” along the dimension of the return volatilities, but less so with regard to the business cycle stylized facts. In particular, investment is insufficiently volatile.

The substance of these theoretical contributions, broadly speaking, is as follows: (1) labor market arrangements have substantial impact on the volatility of profits and shareholder income. (2) In contexts where shareholders have limited ability to hedge this added income risk, its consequences for the equilibrium pricing of financial claims are profound and generally go in the direction of enhancing the models abilities to simultaneously replicate the stylized facts of the business cycle and financial markets. (3) Since the magnitude of the equity premium responds directly to low-frequency income shocks, it is convenient—in the sense of allowing for a superior replication of financial data within a simple context—to have alternative sources of income variation beyond that arising from business cycle co-movements. Our risk sharing mechanism is one such source. (4) A reasonable representation of the financial stylized facts requires income shocks that cannot be insured (smoothed). This may take the form of technological restrictions, as in many of the papers detailed in the present section, or various forms of market incompleteness. Our distribution risk perspective entails aspects of both these perspectives.27

TABLE 16. Model Results: Guvenen (2005)i

Financial and aggregate statistics(a)ii(b)(c)re5.3014.10—rf1.985.73—rp3.3214.700.30(e)Output2.4Total consumption2.3Shareholder consumption4.6Non-shareholder consumption1.1Investment2.7Capital stock—

The focus of research in empirical finance is to explain the cross section of security returns, where the notion of cross section is in reference to sets of specifically constructed portfolios rather than individual issues.28 Curiously, there are to date few studies that include labor market explanatory variables of any sort in the first stages of the Fama and MacBeth (1972) style regressions, which constitute the fundamental technique employed in these exercises. There are two exceptions to this general rule: Jagannathan and Wang (1996) and Santos and Veronesi (2004). Jagannathan and Wang (1996) include the growth rate in per capita labor income as an explanatory variable, a fact that allows their model to outperform the standard CAPM. Such a variable is completely consistent with the model presented in this paper: an above-average value of μ¯tin a particular period is consistent with a simultaneous high growth rate in per worker labor income, and vice versa.

Santos and Veronesi (2004) focus rather on the predictability of stock returns. Their labor market variable is the economy-wide labor income to total consumption ratio, a quantity that is perfectly positively correlated with μtin our model. What is of particular interest to us is the intuition provided by Santos and Veronesi (2004), all of which, by construction, applies to the “distributional risk” construct. They argue that the share of income due to wages, as with all other principal sectors of the economy is a stationary process. The significance of this fact for asset pricing is twofold: (1) if the share of income to labor is high and likely to remain so, investors' MRS variability will be relatively insensitive to events in the stock market, and thus the market risk premium is likely to be small, and vice versa. We note, however, that this ignores the operating leverage effect: a higher share of income to labor suggests a higher fundamental riskiness in the equity cash flow, something, per se, likely to increase the premium. For most all cases presented in this chapter, the latter effect dominates the Santos and Veronesi (2004) intuition. (2) If the share of income from wages is above average, it is likely to decline with the consequence that future dividend growth is likely to exceed consumption growth, leading to high asset prices and returns. The time variation in the asset risk premium suggested by these comments, however, is fully a feature of the distributional risk model.

Regressing stock returns on lagged values of this variable leads to statistically significant coefficients and adjusted R2 that exceed what would be obtained using, e.g., the lagged dividend-price ratio as the explanatory variable. We suspect that a similarly good fit could be obtained using data generated by our model. Including this ratio as an explanatory variable also allows the model to outperform the standard CAPM in explaining the returns to the 25 Fama and French (1993) portfolios. Given the results obtained here, none of this is surprising.

4.4 Impact Evaluation and Future Perspectives

In principle MEA, should be evaluated on different perspectives: organisational and economic impact, contribution in reducing uncertainty, improvement in drugs market uptake, impact on innovation process.

Design and evaluation of outcome-based MEA were discussed for the first time by McCabe and colleagues (McCabe etal.,2010) and Menon and colleagues (Menon etal.,2010). However, most of these first proposals have been systematically analysed by the Ispor Task Force for Performance Based Risk Sharing Arrangements (PBRSA, i.e. outcome-based agreements), with a focus on CED (Garrison etal.,2013).

The Ispor Task Force identified requirements for implementation, governance and evaluation of CED. The Task Force on outcome-based MEA report indicates that evaluation should rely on process indicators for the scheme’s success. ‘It will be an important part of the design of any Performance Based Risk Sharing Arrangements ... to define the metrics by which the success of that scheme can be assessed’ (Garrison etal.,2013). The Task Force identified questions to seize process indicators of success. These questions include: were the intended outcome measures collected? Was uncertainty in associated parameter estimation reduced for the outcomes that were the focus of the scheme? Did the scheme keep within budget? Was the integrity of the design/estimation maintained? Did the governance arrangements work? Did the process to underpin a decision with further evidence work? According to the Task Force ‘The appropriate decision-making will require the ability to show that the agreed outcome adjustments were made to guarantee the cost-effectiveness of the intervention.’

The Task Force has pointed out that a good implementation of CED can be favoured if (i) outcomes are clinically robust, plausible, appropriate and monitorable; (ii) costs of the schemes are acceptable to the health care system and proportionate to the potential gains; (iii) the time horizon is realistic and timeline adequately pre-planned, considering the changing clinical practice and technological advancement. It has also been suggested to provide for (i) a clear provision for funding arrangements and an agreement on responsibility for data collection and analysis; (ii) an agreement on the economic consequences of the CED. These consequences can be defined in advance or left to negotiation between parties; (iii) an exit strategy, in case the agreements cannot be carried out anymore.

The governance of MEA may range from very simple bilateral negotiation for most financial-based contracts to very complex structures for CED, where (i) different stakeholders beyond the industry and payers are involved (researchers, clinicians, patients), and in some instances (ii) steering and scientific committees are usually created to supervise the agreement implementation and its scientific content, respectively.

Transparency is also advocated for reporting in outcome-based agreements, especially for CED carried out only ‘in research’, where coverage is restricted to patients receiving the drug as part of a clinical study or registry (Carlson etal.,2010). The rationale for the publication of the results of an outcome-based MEA is the public nature of these data.

More in general, outcome-based MEA should not involve appreciable administrative burden: for example, already available information systems, such as patients’ registries, could be customised to implement PLR agreements. Furthermore, parties involved should take into consideration possible changes which may impact the agreement, such as new drugs launch or changes in therapeutic guidelines.

The most prominent example of problems arising from governance issues in CED agreements is the risk sharing agreement for β interferons and glatiramer in the treatment of Multiple Sclerosis in UK. In its initial appraisal, the NICE refused to fund these drugs on clinical and cost-effectiveness grounds (McCabe etal.,2010). In 2002, the Government established a CED with the four relevant manufacturers, where a cohort of approximately 5000 patients in 72 centres would be followed for 10years with the maximum cost of drugs funded of £35,000/QALY. The interim analysis, expected after 2years, was published only in 2009 (Boggild etal.,2009), due to problems in recruiting patients. It was observed that the results were not positive with an actual benefit (measured with the Expanded Disability Status Scale) lower that the expected one and the untreated comparator dataset. However, the authors declared that ‘It was too early to reach any conclusion about the cost effectiveness of disease modifying treatments from this first interim analysis. Important methodological issues, including the need for additional comparator datasets, the potential bias from missing data, and the impact of the ‘no improvement’ rule, will need to be addressed and long-term follow-up of all patients is essential to secure meaningful results’ (Boggild etal.,2009). Many researchers advocated for stopping the experiment (among them McCabe etal.,2010; Raftery,2010) or criticised the way the study was designed: for example, Sudlow and Counsell (2003) were very critical for having companies as co-funder and would have preferred a fully independent study. The risks of not having available these treatments anymore within the scheme and a switch of patients to new more expensive drugs, convinced the parties involved continuing the scheme, thus revealing a difference of opinion in the scientific community. Furthermore, the scheme was criticised (Adamski etal.,2010; Towse etal.,2012) because of the delays in patients’ recruitment, contract’s incompleteness, doubts over independence of the Scientific Advisory Group and because hospitals did not receive additional funding for the extensive follow-up consultations and the insufficient infrastructure required including specialist nurses. A new study was initiated with a new natural history dataset and an improved cost-effectiveness model (Palace etal.,2014) and results on 6years evidence were published in 2015 (Palace etal.,2015).

Despite the recommendation by the Ispor Task Force, impact evaluation of MEA is quite rare. Except for studies referring to the UK Multiple Sclerosis CED, other studies included only qualitative discussions of costs and benefits, without evaluating the overall economic impact of the scheme (Puig-Peiró etal.,2011). In the survey conducted by Ferrario and Kanavos (2013) financial and administrative burden estimates, required to responders, were rarely reported, and, if reported, the information was often incomplete. For example, in Italy, only people working on drugs registries in AIFA were reported, but the administrative burden is mostly driven by the time spent by the end-users of registries in hospitals (Garattini etal.,2015). Some studies have provided evidence on (i) administrative costs of PAS in England (Williamson,2010), (ii) reduction in prices due to discontinuation rate of bosentan in Australia (Wlodarczyk etal.,2011), (iii) speeder access of cancer drugs subject to PLR in Italy (Russo etal.,2010), but with a limited application of payback for non-responders (Navarria etal.,2015). Despite the confidentiality nature of the agreements, a higher level of transparency has been advocated at least on their impacts. For example, it has been stressed that, despite the availability of drugs registries in Italy, there are no published report on the actual impact of new drugs except for a summary report in 2007 (Garattini and Casadei,2011).

The impact on the industry has been investigated only theoretically. On the one side, more predictability of price has been considered as a factor that could encourage innovation (Cook etal., 2008; Stafinski etal.,2010), but this may have an impact on the overall investments in Research and Development and not on the location of these investments. In fact, recent evidence shows that cost-containment policies (including unexpected price cut) are not a driver of investments location by the companies (Eger and Mahlich,2014). On the other hand, reallocation of the costs of failures and uncertainty about future income stream (McCabe etal.,2010) might act as a disincentive for the industry.

Notwithstanding the limited evidence on their impacts, MEA have being growing in the last decade and this has facilitated patients access to drugs characterised by an important uncertainty around their clinical and financial impacts. MEA extension to other health technologies is moving its first steps (e.g. medical devices, Fuchs etal.,2017) or is under discussion (e.g. vaccines, Baron-Papillon etal.,2014).

Financial-based contracts are still more frequent than outcome-based contracts, due to their easier management, the prominent importance of the financial impact of new drugs, and the frequent high uncertainty around the target population, which is an important component of the financial impact of high-cost drugs. Among outcome-based contracts most countries have adopted a CED approach, with the relevant exception of Italy, where PLR have been extensively used for cancer drugs. It is unlikely that many countries will adopt a PLR approach, due to the need of a drugs registry.

Future perspectives of MEA (especially outcome-based) depend on the convenience of stipulating them and their ability to face uncertainty issues.

The industry will support MEA if they favour market access. However, the larger the MEA are used by payers the more the industry will incorporate in their price proposals the expected impact of these contracts.

Payers will support MEA if their pros (reducing the financial impact of new drugs, allowing market access, collecting real world data) will compensate the cons (asymmetric information at contract sign, management costs, increase in prices requests). Health care professionals and patients will take advantage of drugs availability, but the professionals will be required additional burden to manage these agreements.

More specifically MEA could be considered in the future, for new treatments in a high priority disease area with an expected net health gain, but with important uncertainty about their effects.

Hence, it is not excluded that MEA might experience further evolution in the coming years, due to the increasing pressure for better health and right price, and the growing dynamics of interrelated drug markets.

What type of contract is the best risk sharing arrangement?

What is a risk arrangement?

Is risk sharing a function of insurance?

What is risk sharing as it pertains to health insurance quizlet?