Contract Nonperformance Risk and Ambiguity in Insurance Markets

Transcription

1 Contract Nonperformance Risk and in Insurance Markets Christian Biener, Martin Eling (University of St. Gallen) Andreas Landmann, Maria Isabel Santana (University of Mannheim) 11 th Microinsurance Conference 1/20

2 Motivation - Insurance policies meant to protect against economic losses - But policies sometimes fail to perform: - Insolvencies - Discord about the losses covered - Payment delays - Exclusions of certain losses - Fraud Leads to contract nonperformance risk - More relevant in emerging and poorly regulated markets 2/20

3 Illustration: Classical Case - Literature focuses on risk, i.e. probabilities can be assigned to all possible outcomes: Claim with probability p Individual Payout for sure Insurer 3/20

4 Illustration: Default Risk - In case there is a default probability: Claim with probability p Individual Payout with probability 1-r Insurer - Might affect demand and market development - probabilistic insurance unattractive for risk-averse (Doherty and Schlesinger 1990) - Similar to downside risk in index insurance (Clarke 2011) - empirically reduces demand (Wakker, Thaler & Tversky 1997) (Herrero, Toms & Villar 2006) (Zimmer, Schade & Gründl 2009) 4/20

5 Illustration: - But probabilities might in reality be unknown, i.e. ambiguous (Epstein,1999) Claim with probability p? Individual Payout with probability 1-? 1-r Insurer - again maybe higher in emerging/unregulated markets - Research on ambiguous loss risk exists (Alary, Gollier & Treich 2013) (Hogarth & Kunreuther 1989) But no work on contract nonperformance ambiguity so far 5/20

6 Contribution r = 0 r > 0 r =? p > 0 Theory: Empirics: Theory: Empirics: Theory: Empirics: p =? Theory: Empirics: Theory: Empirics: Theory: Empirics: Theory We extend insurance theory by introducing ambiguity of contract nonperformance risk (r) We interact contract nonperformance risk with ambiguous shock probabilities (p) Empirics We validate our theoretical results in a particularly relevant field-lab setting: an emerging insurance market (Philippines) We test the impact of ambiguous contract nonperformance risk We test the interaction between deterministic contract nonperformance risk and ambiguous shock probabilities We test for framing effects We test for updating processes in a dynamic setting 6/20

7 Theory (1) - We use von Neumann-Morgenstern expected utility - is implied by a set of possible loss or nonperformance probabilities p γ and r(γ) - aversion is introduced using a concave valuation function for expected utility under different states of γ (Klibanoff, Marinacci & Mukerji 2005) aversion essentially makes individuals more pessimistic about probabilities p and r 7/20

8 Illustration Theory (2) - We proof several Lemmas, from which our main hypotheses follow: H1: Contract nonperformance risk reduces insurance demand H2: about contract non-performance probabilities reduces insurance demand for ambiguity-averse individuals H3: about loss probabilities increases insurance demand for ambiguity-averse individuals - Additionally: H4: Negatively framing contract nonperformance risk decreases demand for probabilistic insurance We test these hypotheses via a behavioral experiment 8/20

9 Basic idea of behavioral experiments - Create decision problem related to the research question - Observe individuals in a controlled environment - Incentivize the decision problem (with money usually) - Vary the environment to observe behavior under different settings Allows observing details which are otherwise hard to observe Allows simulation of policy experiments External validity? 9/20

10 Sample - Lab-in-the-field experiments in 42 rural villages from central Philippines - Two-stage sampling: - Villages drawn randomly from two provinces - Within villages, representatives from random households invited experimental sessions conducted (four in each village) - In total 996 participants (six per session) 10/20

11 Geographic Location Sample Characteristics (N=996) Sulu Sea Panay Gulf Iloilo Variable Mean SD Age (in years) Gender (1=female) 0.79 Years of education Married a 0.90 Employed a 0.40 Regular income a 0.26 Seasonal income a 0.72 Household owns land a 0.14 Skipped meals in last month a a Indicator variable where 1 is yes and 0 is no. 11/20

12 Experimental Procedure Time line Pre-experimental survey Risk/ambiguity preferences elicitation Explanation of the game and comprehension checks Procedures for one game round (6 rounds played): Cash Flows 1Initial endowment. 2Decision about insurance uptake.. 3Draw shock... 4Draw claim payment if shock & insurance.. Post-experimental survey and payout /-60/0-150/0 150/0 12/20

13 Generating - Make loss and nonperformance probabilities unknown - At the same time give them an average idea about probabilities - Basic idea: (10 grey / 90 blue) (? grey /? blue) - Similar for ambiguity w.r.t. loss probability - Creates common knowledge about distribution of probabilities 13/20

14 Insurance Uptake Notes: Average uptake across six rounds, confidence intervals are on 95% level, standard errors account for clustering of standard errors at the session level. Control No Default Default Loss Negative Default Framing Default & Negative Framing 14/20

15 Average Treatment Effects H1 Hypotheses Contract nonperformance risk reduces insurance demand H2 H3 H4 about contract nonperformance probabilities reduces insurance demand about loss probabilities increases insurance demand Negatively framing contract nonperformance risk decreases demand for probabilistic insurance?? Note: Standard errors in parentheses are clustered at the session level. a Probit model results in terms of marginal effects. 15/20

16 Further Results - Results by numeracy: - Effects larger and more significant in high numeracy subsample - Effect of loss ambiguity more positive with high numeracy, but still insignificant - Little significant effects in low numeracy sample - Results by ambiguity aversion: - averse subjects react more to presence of ambiguity Table Table 16/20

17 Results over time Default Loss Default & Negative Framing Effect on uptake 0.2 r1 r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 Suggests that ambiguity is not easily resolved 17/20

18 Probability guesses over time best guess min guess max guess Round Round Round Default Risk Default Default Risk Default Default Risk Default Default & Negative Framing Loss & Negative Framing Default & Negative Framing & Negative Framing 18/20

19 Probability guesses over time: Spreads and Errors max - min mean - real Default Risk Round Round Default Default & Negative Loss Framing Round Default Risk Default & Negative Framing Default & Negative Framing 19/20

20 Discussion - Contract nonperformance risk and ambiguity may play a significant role for demand in emerging insurance markets - not easy to resolve - Limited number of claims observable - Updating beliefs nontrivial - Regulatory environment important - Ensuring low levels of contract nonperformance - Limit ambiguity by increasing market transparency - Argument for insurers to focus on sound policies and to build trust in the market 20/20

21 Thank you for your attention 21/20

22 Back Effect of nonperformance and loss ambiguity - Total effect depends on the joint distribution of risk and ambiguity aversion (Vieider et al. 2015) Simulations with ambiguous nonperformance: Simulations with ambiguous loss: 22/20

23 Panel A: Universal Parameters Initial endowment: 210 PHP Loss: 150 PHP Loss prob. prior Panel B: Treatment characteristics Nonperformance prob. prior Ambiguous loss prob. Ambiguous nonperformance prob. Framing Premium Control 30% 10% neutral 50 PHP TNoDef 30% 0% neutral 60 PHP TDef 30% 10% neutral 50 PHP TLoss 30% 10% neutral 50 PHP CFr 30% 10% negative 50 PHP TDef-Fr 30% 10% negative 50 PHP 23/20

24 Average Treatment Effects by Numeracy H1 H2 H3 H4 Hypotheses - Effects more pronounced in high numeracy subsample - Effect of loss ambiguity more positive with high numeracy, but still insignificant - Little significant effects in low numeracy sample Note: Standard errors in parentheses are clustered at the session level. p < 0.1, p < 0.05, p < 0.01 significance level at 10, 5 and 1 percent. Back 24/20

25 Average Treatment Effects by Aversion averse subjects react more to presence of ambiguity Note: Standard errors in parentheses are clustered at the session level. p < 0.1, p < 0.05, p < 0.01 significance level at 10, 5 and 1 percent. Back 25/20

26 over Time Updating Process Information on ambiguous probabilities accumulate through experience Own Experience Network Experience Rational individuals should update beliefs about the unknown stochastic process With more observations, the true probability can be estimated more precisely Hypothesis Subjective probability distribution over the possible probabilities should converge towards true probability With decreasing ambiguity the treatment effects of ambiguity converge to zero 26/20