APPLICATION OF MONTE CARLO SIMULATION IN INSURANCE RISK ASSESSMENT

CHAPTER ONE

INTRODUCTION

1.1 Background to the Study

The central quantitative problem in general insurance is the determination of the aggregate loss distribution the probability distribution of total claims arising from a portfolio over a period. From this distribution follow the quantities that govern the solvency of an insurer: the expected loss, the premium required, the capital needed to survive adverse outcomes at a chosen confidence level, and the probability of ruin.

Analytical approaches to this problem are limited. The compound distribution formed from a claim count distribution and a claim severity distribution rarely has a closed form. Panjer recursion offers an exact solution for certain count distributions but requires discretisation of severity and becomes computationally awkward for large portfolios or complex reinsurance structures. Monte Carlo simulation avoids both constraints. By repeatedly sampling a claim count from the fitted frequency distribution, then sampling that many claim amounts from the fitted severity distribution, and summing, the method generates an empirical aggregate loss distribution of arbitrary precision. It accommodates dependence between risks, non-linear reinsurance structures, multi-year projection and stochastic investment return with no change of method.

The approach is well established in actuarial practice. Under Solvency II, partial internal models for large claims are routinely built on Monte Carlo simulation of aggregate losses, from which the capital requirement is derived as a high quantile of the simulated loss distribution. Euphasio Junior and Carvalho (2022) simulated the Cramér–Lundberg risk process by Monte Carlo across thirty lines of business calibrated on nearly four million claim records, estimating ruin probabilities at varying solvency capital levels with and without quota share and excess of loss reinsurance. Comparative work has shown that the choice of severity distribution materially affects the estimated ruin probability, with heavy-tailed specifications such as the lognormal producing substantially higher ruin probabilities than the exponential at the same capital level.

The Nigerian context makes this methodology newly relevant. The Nigerian Insurance Industry Reform Act, 2025 introduced risk-based capital, requiring insurers to hold capital calibrated to their specific insurance, market, credit and operational risks, together with a capital adequacy ratio of 100 per cent (Federal Republic of Nigeria, 2025). Risk-based capital cannot be determined by rule of thumb; it requires a probability distribution of loss. Yet Nigerian insurers face a claims environment with pronounced tail risk oil and gas exposures accounting for 30.3 per cent of non-life premium, flooding, building collapse and fire alongside an actuarial capacity deficit that limits the deployment of sophisticated internal models. Simulation is attractive here precisely because it is conceptually simple and computationally cheap, requiring no closed-form mathematics and running adequately on standard hardware in R or Python. This study develops and applies a Monte Carlo simulation

framework for insurance risk assessment using Nigerian portfolio data.

1.2 Statement of the Problem

Nigerian insurers are required to hold risk-based capital but have no established methodology for computing the loss distribution from which such capital would be derived. Four problems follow.

First, deterministic reserving in a stochastic environment. Reserves and capital in Nigerian practice are commonly set as multiples of premium or by reference to prescribed minima. Neither approach produces a probability statement about the adequacy of the reserve, which is what risk-based supervision demands.

Second, tail risk is invisible under average-based methods. The capital requirement is determined by the extreme quantiles of the loss distribution, not the mean. An insurer pricing off average claim experience holds no information about the 99.5th percentile outcome that determines whether it survives a bad year.

Third, the effect of reinsurance is not quantified. Nigerian insurers purchase quota share and excess of loss protection, but the effect of a given reinsurance structure on the retained loss distribution, on the ruin probability and on the required capital is not routinely computed. Without this, reinsurance purchase is a negotiation rather than an optimisation.

Fourth, no Nigerian application exists. The simulation literature is developed almost entirely on European, North American and Latin American data. Whether Nigerian claims data with their heavier tails, higher inflation adjustment requirements and shorter reliable histories support stable simulation-based estimates has not been tested.

The problem, therefore, is the absence of a validated Monte Carlo framework for assessing insurance risk in the Nigerian market at the moment when risk-based capital regulation presumes one exists.

1.3 Aim and Objectives of the Study

The aim of this study is to apply Monte Carlo simulation to insurance risk assessment using Nigerian insurance portfolio data.

The specific objectives are to:

1. fit appropriate frequency and severity distributions to the claims data of the study portfolio;

2. develop a Monte Carlo simulation model of the aggregate claim distribution for the portfolio;

3. estimate risk measures expected loss, value at risk and expected shortfall from the simulated aggregate loss distribution;

4. estimate the probability of ruin at varying levels of initial capital using a simulated Cramér–Lundberg risk process;

5. evaluate the effect of alternative reinsurance structures on the retained loss distribution, the required capital and the ruin probability; and

6. assess the convergence and stability of the simulation estimates as the number of replications increases.

1.4 Research Questions

1. Which frequency and severity distributions best fit the claims data of the study portfolio?

2. What aggregate loss distribution results from Monte Carlo simulation of the fitted models?

3. What values of expected loss, value at risk and expected shortfall are implied by the simulated distribution?

4. What is the probability of ruin at varying levels of initial capital?

5. How do alternative reinsurance structures affect the retained loss distribution, the required capital and the ruin probability?

6. How many replications are required for the simulation estimates to converge acceptably?

1.5 Research Hypotheses

The following null hypotheses will be tested at the 5% level of significance:

H₀₁: There is no significant difference between the aggregate loss distribution obtained by Monte Carlo simulation and that obtained by analytical or recursive approximation.

H₀₂: The choice of severity distribution has no significant effect on the estimated ruin probability.

H₀₃: The level of initial capital has no significant effect on the probability of ruin.

H₀₄: Reinsurance arrangements have no significant effect on the required solvency capital of the portfolio.

1.6 Significance of the Study

For insurance companies, the study supplies a practical, implementable methodology for determining risk-based capital, testing reinsurance structures and quantifying the probability of insolvency capabilities that NIIRA 2025 now effectively requires. For NAICOM, it demonstrates a computational approach to internal model validation and offers a benchmark against which insurers' submitted capital calculations can be reviewed. For reinsurers and reinsurance brokers, the quantified effect of alternative treaty structures on the ceding company's loss distribution supports evidence-based programme design and pricing. For actuarial students and practitioners in Nigeria, the study provides a worked, locally calibrated application of simulation methodology at a time when the profession is being asked to expand rapidly. For scholarship, it extends the simulation literature into a sub-Saharan African market whose data characteristics heavy tails, high inflation, short reliable histories differ materially from those on which the methods were developed, thereby testing their robustness.

1.7 Scope of the Study

The study covers claims data from one or more general insurance portfolios of a Nigerian insurance company licensed by NAICOM, over a period of not less than five consecutive years. The classes examined are those with sufficient claim volume for distribution fitting, typically motor, fire and general accident, with oil and gas included where data permit given its dominance of the market. The methodological scope covers distribution fitting by maximum likelihood, goodness-of-fit testing, Monte Carlo simulation of the compound aggregate loss distribution, simulation of the Cramér–Lundberg surplus process, estimation of value at risk and expected shortfall, and the simulation of quota share and excess of loss reinsurance structures. Implementation is in R or Python. The study does not extend to life insurance liabilities, to nested stochastic asset–liability projection, or to the calibration of market, credit and operational risk components of the capital requirement.

1.8 Limitations of the Study

(i) Data quantity reliable simulation requires enough historical claims to fit severity distributions credibly in the tail, and Nigerian portfolios may contain too few large claims for stable tail estimation. (ii) Inflation adjustment claim amounts must be restated to a common price basis before fitting, and with inflation above thirty per cent during part of the study period, results are sensitive to the deflator chosen. (iii) Parameter uncertainty the simulation treats fitted distribution parameters as known, whereas they are estimated with error; unless parameter uncertainty is itself simulated, risk measures will be understated. (iv) Independence assumptions the basic model assumes claims are independent and identically distributed, which catastrophe events such as flooding and civil disturbance violate. (v) Computational resources high-quantile estimation requires large numbers of replications, which may exceed available computing capacity. (vi) Data access insurers may release only aggregated claims data, which would prevent individual-claim severity fitting.

1.9 Operational Definition of Terms

Monte Carlo simulation: A computational technique that estimates the distribution of an output quantity by repeatedly sampling from the probability distributions of its inputs.

Aggregate loss distribution: The probability distribution of total claims arising from a portfolio over a period, formed as a compound distribution of claim frequency and claim severity.

Compound Poisson process: A stochastic process in which claim arrivals follow a Poisson process and claim amounts are independent, identically distributed random variables.

Cramér–Lundberg model: The classical collective risk model in which an insurer's surplus evolves as initial capital plus premium income less aggregate claims.

Probability of ruin: The probability that an insurer's surplus falls below zero at some point within a defined horizon.

Value at Risk (VaR): The loss level that will not be exceeded with a specified probability over a specified period.

Expected shortfall (conditional tail expectation): The expected loss conditional on the loss exceeding the value at risk threshold.

Solvency capital requirement: The capital an insurer must hold to limit the probability of insolvency over a defined horizon to a specified level.

Quota share reinsurance: A proportional treaty under which the reinsurer accepts a fixed percentage of every risk and of every claim.

Excess of loss reinsurance: A non-proportional treaty under which the reinsurer pays the portion of a claim exceeding an agreed retention, up to an agreed limit.

Convergence: The stabilisation of a simulation estimate as the number of replications increases.

References

Artzner, P., Delbaen, F., Eber, J. M., & Heath, D. (1999). Coherent measures of risk. Mathematical Finance, 9(3), 203–228. https://doi.org/10.1111/1467-9965.00068

Asmussen, S., & Albrecher, H. (2010). Ruin probabilities (2nd ed.). World Scientific.

Euphasio Junior, J. W., & Carvalho, J. V. F. (2022). Reinsurance and solvency capital: Mitigating insurance companies' ruin probability. Revista de Administração Contemporânea, 26(1), e200191. https://doi.org/10.1590/1982-7849rac2022200191.en

Federal Republic of Nigeria. (2025). Nigerian Insurance Industry Reform Act, 2025. Federal Government Press.

Glasserman, P. (2004). Monte Carlo methods in financial engineering. Springer.

Kaas, R., Goovaerts, M., Dhaene, J., & Denuit, M. (2008). Modern actuarial risk theory: Using R (2nd ed.). Springer.

Klugman, S. A., Panjer, H. H., & Willmot, G. E. (2019). Loss models: From data to decisions (5th ed.). John Wiley & Sons.

Korn, R., Korn, E., & Kroisandt, G. (2020). A guide to Monte Carlo simulation concepts for assessment of risk-return profiles for regulatory purposes. European Actuarial Journal, 10(2), 273–293. https://doi.org/10.1007/s13385-020-00232-3

National Insurance Commission. (2026). Bulletin of the insurance market performance: Fourth quarter 2025. NAICOM.