INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Determinants of Surrenders in Life Insurance: Evidence from

Tunisian Periodic Savings

Mehrez Ben Nasr

Certified Actuary (FTUSA)

DOI: https://dx.doi.org/10.51244/IJRSI.2025.1210000269

Received: 02 November 2025; Accepted: 08 November 2025; Published: 18 November 2025

ABSTRACT

This paper investigates the determinants of policyholder surrenders in Tunisian periodic-savings life insurance

using real-world data extracted from a Tunisian insurance company. A logistic regression model identifies the

main behavioral and financial drivers of surrender decisions in an emerging North African market context. The

analysis shows that prior partial surrender (odds ratio ≈ 6.25), the existence of policy advances (≈ 1.82), and low

mathematical reserves (< 80 KDT; ≈ 0.56) are the key explanatory variables. The model achieves an AUC of

0.96–0.97, indicating high predictive accuracy comparable to advanced machine learning approaches reported

in developed markets. These findings provide actionable insights for life insurers in emerging economies,

supporting targeted retention strategies, improved liquidity planning, and enhanced capital management. The

study contributes to the limited empirical literature on surrender behavior in African insurance markets and

validates key theoretical predictions from international research in the Tunisian context.

Keywords: Life Insurance, Surrender, Tunisia, Logistic Regression, Actuarial Analysis, Emerging Markets,

North Africa

INTRODUCTION

Policyholder surrender represents a critical challenge for life insurers worldwide, directly influencing

profitability, liquidity, and solvency. In emerging markets such as Tunisia, understanding surrender behavior is

essential to anticipate cash flows and maintain capital adequacy under evolving regulatory frameworks. The

Tunisian insurance market, despite being the fourth largest in the MENA region with a market penetration of

2.7%, faces persistent challenges in life insurance development, where life products represent only 17% of total

premiums. This structural imbalance underscores the importance of understanding policyholder behavior,

particularly surrender decisions that can undermine the long-term viability of life insurance operations.

Surrender risk has gained renewed attention globally following the interest rate environment shifts of recent

years. Life insurers face the dual challenge of managing liquidity risk from mass surrenders while maintaining

adequate returns for policyholders. The surrender option embedded in most life insurance contracts creates an

asymmetric risk profile: when market interest rates rise above guaranteed rates, policyholders have strong

economic incentives to surrender their policies and reinvest elsewhere. This phenomenon, documented

extensively in developed markets, remains largely unexplored in African and Middle Eastern contexts where

market structures, regulatory environments, and policyholder behaviors may differ substantially.

This research addresses three key gaps in the existing literature. First, while extensive research examines

surrender behavior in developed markets, empirical evidence from North African and Middle Eastern insurance

markets remains scarce. Second, most studies focus on macroeconomic determinants of aggregate surrender

rates, with limited attention to individual contract-level characteristics that predict surrender at the policyholder

level. Third, emerging market studies often lack the methodological rigor and model performance metrics

necessary for practical implementation in insurer risk management systems.

The present study contributes by offering empirical evidence from Tunisia, an emerging market with unique

economic, demographic, and behavioral characteristics. Using a comprehensive dataset of periodic-savings

Page 3104

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

contracts obtained from a real-life Tunisian insurance company (data details remain confidential for privacy and

proprietary reasons), this research identifies key determinants of surrender and provides professional

recommendations for insurers seeking to manage these risks effectively.

In addition, and following prior methodological work (Ben Nasr, 2025, International Journal of Research

Publication and Reviews), alternative modeling approaches—such as Decision Trees, Support Vector Machines

(SVM), and Neural Networks—were tested using the same real-world dataset. The comparative analysis

demonstrated that logistic regression achieved the highest predictive performance (AUC = 0.96–0.97) while

maintaining superior interpretability and calibration stability. Therefore, the logistic regression model was

retained as the core analytical framework in the present study, combining both predictive accuracy and actuarial

transparency.

The findings have important implications for product design, pricing, reserving, and regulatory capital

requirements under evolving solvency frameworks.

LITERATURE REVIEW

A. Theoretical Framework of Surrender Behavior

The theoretical foundation for understanding life insurance surrenders derives from multiple streams of research.

The emergency fund hypothesis posits that policyholders surrender policies primarily to meet unexpected

liquidity needs arising from adverse life events such as unemployment, divorce, or health shocks. Empirical

support for this hypothesis comes from panel data studies showing that household-level shocks significantly

increase surrender probability. The interest rate hypothesis, in contrast, emphasizes that surrenders reflect

rational economic behavior in response to changes in the opportunity cost of maintaining insurance policies.

When market interest rates exceed policy crediting rates, the surrender option moves 'into the money,' creating

strong economic incentives to terminate contracts.

A third theoretical perspective emphasizes behavioral factors including lack of knowledge about policy features,

social influence from family and agents, and company-related service quality issues. This behavioral dimension

is particularly relevant in emerging markets where financial literacy may be limited and insurance penetration

remains low. The mis-selling hypothesis suggests that many surrenders result from policies that were

inappropriate for policyholders' needs at inception, often due to aggressive agent incentives.

Recent theoretical advances recognize that surrender decisions involve multiple, interacting factors operating

simultaneously. Fang and Kung (2012) develop a dynamic structural model incorporating income shocks, health

status changes, and bequest motive variations, demonstrating that the relative importance of these factors varies

systematically with policyholder age. Their findings suggest that i.i.d. choice-specific shocks dominate surrender

decisions for younger policyholders, while income, health, and bequest motive shocks become increasingly

important as policyholders age.

B. Empirical Evidence from Developed Markets

The empirical literature on life insurance surrender is extensive in developed markets. Macroeconomic studies

consistently find that aggregate surrender rates respond positively to increases in market interest rates. Russell

et al. (2013) analyze U.S. industry data and document positive relationships between surrender activity and

inflation, real interest rates, and unemployment, while finding an inverse relationship with household liquidity.

Kuo et al. (2003) and Kiesenbauer (2012) report similar findings for Taiwan and Germany, respectively.

Recent European research provides compelling causal evidence. Kubitza et al. (2023) exploit plausibly

exogenous variation in monetary policy to identify a causal effect of interest rates on surrender rates, addressing

concerns about omitted variable bias in earlier correlational studies. They estimate that a 25 basis point annual

increase in interest rates would force insurers to sell approximately 2% of their investment portfolios to fund

surrender payouts, with significant implications for financial stability. This research highlights life insurance

Page 3105

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

convexity—the property that contract duration decreases with rising interest rates, opposite to the behavior of

most fixed-income securities.

Microeconomic studies using contract-level or household-level data provide richer insights into individual

surrender decisions. Eling and Kochanski (2013) conduct a comprehensive review identifying key product

features affecting lapse including surrender charges, bonus structures, premium payment frequency, and policy

duration. They find that policies with significant biometric insurance components and substantial surrender

penalties experience materially lower lapse rates. Gemmo et al. (2018) use German socioeconomic panel data to

demonstrate that life events imposing liquidity shocks—including childbirth, divorce, dwelling acquisition, and

unemployment—significantly increase surrender probability.

C. Methodological Approaches to Surrender Prediction

The methodological literature on surrender prediction has evolved considerably. Traditional actuarial approaches

rely on generalized linear models (GLMs), particularly logistic regression for binary surrender outcomes. These

methods offer interpretability, regulatory acceptance, and computational efficiency, making them the industry

standard for many insurers. Boonmeekham et al. (2019) demonstrate that logistic models incorporating age, face

amount, payment duration, and occupation can achieve 67% accuracy in predicting individual policy lapses.

Recent studies increasingly apply machine learning techniques including random forests, gradient boosted

machines (GBM), XGBoost, support vector machines (SVM), and neural networks. Kiermayer et al. (2021)

provide extensive experimental comparisons showing that XGBoost and random forests outperform GLMs in

predictive accuracy, though they caution that resampling techniques commonly used to handle class imbalance

can introduce significant probability bias. Azzone et al. (2022) demonstrate that random forests achieve superior

classification performance while maintaining reasonable explainability through SHAP values.

The literature reveals an important trade-off between predictive accuracy and probability calibration. While

machine learning models often achieve higher AUC scores and better classification of surrender versus

persistence, they may produce biased probability estimates unless carefully calibrated. For regulatory capital

calculations and cash flow projections, well-calibrated probabilities are essential, potentially favoring simpler

GLM approaches. Loisel et al. (2019) propose transforming classification problems into regression frameworks

with subsequent optimization, demonstrating significant economic gains.

Model performance evaluation in the literature typically employs multiple metrics including AUC-ROC, log-

loss, precision-recall curves, calibration charts, and lift curves. Studies report AUC values ranging from 0.70 to

0.97 depending on data quality, variable richness, and modeling approach. The highest reported performance

comes from ensemble methods combining multiple algorithms.

D. Surrender Behavior in Emerging Markets

Research on surrender behavior in emerging markets remains limited but growing. Asian markets have received

the most attention. Studies from India reveal that inadequate knowledge about insurance products, social

influence from agents, and tax-saving motivations (rather than protection needs) lead to high lapse rates

exceeding 40% in some segments. Research from China documents similar patterns, with surrender rates

strongly influenced by agent behavior and policy design features.

The limited research on African insurance markets focuses predominantly on market development challenges

rather than policyholder behavior. The Tunisian insurance market, while relatively developed within Africa,

exhibits characteristics typical of emerging markets: low penetration (2.7% of GDP), concentrated market

structure with 23 domestic companies, and life insurance representing only 17% of total premiums. Political

instability since 2011 and regional security concerns add additional complexity to the operating environment.

Recent work from Nepal by Ghimire et al. (2024) provides the most directly relevant emerging market evidence

for this study. Using survey data from 445 policyholders who surrendered during the COVID-19 pandemic, they

identify economic factors as the primary driver, followed by knowledge gaps, social influences, and company-

related factors. Their finding that demographic variables show no significant association with surrender reasons

Page 3106

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

suggests that behavioral and financial factors dominate over traditional sociodemographic predictors in emerging

market contexts.

E. Gap Analysis and Research Contribution

This review reveals several important gaps that the present study addresses. First, geographic coverage remains

heavily skewed toward developed markets, with virtually no empirical research on North African insurance

markets. Second, most emerging market studies rely on survey methods rather than comprehensive contract-

level administrative data, limiting the precision and generalizability of findings. Third, few studies report model

performance metrics (particularly AUC) necessary for benchmarking and practical implementation.

The present research contributes by: (1) providing the first contract-level analysis of surrender determinants in

Tunisia, filling a significant geographic gap; (2) employing rigorous logistic regression methodology with

comprehensive performance evaluation; (3) achieving model performance (AUC 0.96-0.97) comparable to or

exceeding advanced machine learning approaches in developed markets; (4) identifying specific contract

features (prior partial surrender, policy advances, mathematical reserves) as key predictors, offering actionable

insights for product design and risk management; and (5) contextualizing findings within the broader theoretical

framework established in developed markets while acknowledging emerging market specificities.

METHODOLOGY

A. Data and Sample Description

Data source and confidentiality:

The analysis relies on administrative policy records from a real-life Tunisian life insurer offering periodic-

savings products. All records were anonymized prior to access, and commercially sensitive information

(company identifiers and exact counts) is withheld to respect privacy and proprietary constraints. Consistency

checks were performed across actuarial and accounting systems before analysis.

Population and observation window:

The dataset covers policies observed between 2010 and 2024. Periodic-savings policies combine regular

premium contributions with guaranteed benefits at maturity and an embedded surrender option, and they

represent a dominant life-insurance product in Tunisia.

Inclusion and exclusion criteria:

We included contracts that

1.

2.

were in force at any point during 2010–2024 or experienced a full surrender in that window; and

( had complete transaction histories (premiums, advances/loans, reserve evolution).

Longitudinal structure:

Contract records contain both static inception attributes (initial sum assured, age at entry, payment frequency,

contractual term) and time-varying fields (mathematical reserves, partial surrenders, policy advances/loans,

attained duration). This panel-style structure allows us to trace trajectories and identify behaviours preceding

surrender.

Outcome definition:

Full Surrender (binary) equals 1 when a policyholder initiates contract termination and receives the surrender

value within the observation window; it equals 0 for contracts remaining in force through their last observation.

Death claims, maturities, and insurer-driven cancellations for non-payment are not coded as surrenders.

Page 3107

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Primary explanatory variables.

 Prior Partial Surrender (binary). Indicator that the policyholder previously withdrew part of the

accumulated value while keeping the contract active (behavioural persistence / liquidity signal).

 Policy Advance/Loan (binary). Indicator of any outstanding loan secured by policy value (liquidity stress

proxy).

 Mathematical Reserves (continuous). Present value of future benefits minus the present value of future

premiums. For interpretability and to reflect the empirical distribution, we also analyse a low-reserve

flag (< 80,000 TND) as a thresholded variant in some specifications.

Control variables:

Candidate controls included policyholder age, gender, attained duration, initial sum assured, premium amount,

and payment frequency. Preliminary diagnostics (association tests and multicollinearity checks) showed limited

additional explanatory power once the three primary variables were included; to maintain parsimony and

interpretability, these controls were dropped from the final specification.

Data quality and preprocessing:

We applied (i) field-level validation (ranges, type checks), (ii) de-duplication of contract IDs, (iii) reconciliation

of surrender status with cash-value transactions, (iv) treatment of rare missing fields via simple imputation for

non-critical descriptors (no imputation for the outcome), and (v) robust handling of legitimate high values (e.g.,

large reserves) without winsorization, as these reflect genuine portfolio heterogeneity. For alternative

specifications sensitive to scaling, continuous variables were standardized on the training split to prevent

leakage.

Rationale for variable coding:

Binary indicators for prior partial surrender and policy advances capture salient, actionable behaviours linked to

liquidity needs. The reserve level summarizes contract maturity and accumulated value—key economic drivers

of the surrender decision—while the thresholded version (<80,000 TND) offers policy-friendly interpretation

and facilitates risk-segmentation use cases.

B. Statistical Model Specification

Logistic regression was selected as the main analytical framework due to its interpretability, stability, and

regulatory acceptance within actuarial practice. Although alternative techniques such as decision trees, ensemble

models, or survival analysis can capture non-linear or temporal effects, comparative experiments conducted in

a separate methodological study (Ben Nasr, 2025, International Journal of Research Publication and Reviews)

showed that logistic regression achieved the highest predictive accuracy (AUC = 0.96–0.97) and calibration

consistency. Survival analysis was not implemented in this study because the available data did not include

precise event-time information required for duration-based estimation. The logistic approach thus offers the

optimal balance between explanatory clarity and predictive performance in this empirical setting.

The analysis employs binary logistic regression, the standard approach for modeling dichotomous outcomes in

insurance applications. The model estimates the probability that contract i experiences surrender as:

P(Yi = 1|Xi) = exp(β₀ + β₁Xi) / (1 + exp(β₀ + β₁Xi))

Equivalently, the log-odds (logit) transformation provides a linear relationship:

logit(p) = ln(p/(1-p)) = β₀ + β₁Xi

Page 3108

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Where Yi is the binary surrender indicator for contract i, Xi is the vector of explanatory variables, βi is the vector

of regression coefficients estimated by maximum likelihood, and pi is the surrender probability for contract i.

The model coefficients β are estimated using maximum likelihood estimation (MLE), which finds parameter

values maximizing the likelihood of observing the actual surrender outcomes in the sample. Statistical

significance is assessed using Wald tests, with conventional significance levels (α = 0.05).

The odds ratio for predictor j is calculated as OR = exp(βj), representing the multiplicative change in surrender

odds for a one-unit increase in the predictor, holding other variables constant. Odds ratios provide intuitive

interpretability: OR > 1 indicates increased surrender risk, OR < 1 indicates decreased risk, and OR = 1 indicates

no effect.

C. Model Validation and Performance Evaluation

Model performance is evaluated using multiple complementary metrics following best practices in the actuarial

and machine learning literature:

 Area Under the ROC Curve (AUC-ROC): The primary performance metric, measuring the model's

ability to discriminate between surrendered and persistent contracts across all possible classification

thresholds. AUC ranges from 0.5 (random discrimination) to 1.0 (perfect discrimination), with values

above 0.80 generally considered excellent. The ROC curve plots true positive rate (sensitivity) against

false positive rate (1-specificity).

 Classification Accuracy: The proportion of contracts correctly classified as surrendered or persistent

using an optimal probability threshold (typically 0.5 or determined by maximizing Youden's index).

 Sensitivity and Specificity: Sensitivity (true positive rate) measures the proportion of actual surrenders

correctly identified, while specificity (true negative rate) measures the proportion of persistent contracts

correctly identified. The optimal model balances both metrics.

 Calibration Assessment: While not formally quantified in this study, model calibration—the

correspondence between predicted probabilities and observed surrender frequencies—was assessed

visually through calibration plots.

The model is validated using holdout validation, randomly splitting the dataset into training (70%) and test (30%)

subsets. Model coefficients are estimated on the training data and performance metrics calculated on the

independent test data to assess generalization capability. This approach provides more conservative performance

estimates than resubstitution (testing on training data) while being computationally more efficient than k-fold

cross-validation for large datasets.

The achieved AUC (0.96-0.97) is compared against performance benchmarks reported in the surrender

prediction literature. Studies using logistic regression on comprehensive datasets typically report AUC values

between 0.70 and 0.85, while advanced machine learning approaches achieve AUC values between 0.85 and

0.95. The present model's performance approaching 0.97 suggests either exceptional discriminatory power of

the selected predictors or relatively clean separation between surrendered and persistent contracts in the Tunisian

data.

D. Comparison with Literature Benchmarks

Methodological comparisons reveal several relevant benchmarks from both developed and emergingmarkets.

Boonmeekham et al. (2019) applied logistic regression to predict life policy surrender in Thailand and achieved

an

Similarly, Shamsuddin et al. (2025) in Malaysia reported a logistic regression accuracy of 79.6%, emphasizing

the model’s interpretability and regulatory suitability in Southeast Asian contexts.

AUC

of

approximately

0.70

with

67%

classification

accuracy.

In contrast, Azzone et al. (2022) in Italy found that Random Forest algorithms substantially outperformed logistic

regression, highlighting the benefits of non-linear modeling for large and complex datasets.

Page 3109

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Kgare (2021) in South Africa obtained 92% accuracy using gradient boosting and 76% with random forests,

suggesting that ensemble methods can yield strong predictive performance in African market settings.

Kiermayer et al. (2021) further demonstrated through simulation studies that XGBoost and random forest models

outperform logistic regression in predictive precision but often suffer from calibration instability and

interpretability issues.

These international and regional benchmarks contextualize the present study’s exceptional AUC performance of

0.96–0.97, which surpasses most previously reported results across both developed and emerging markets.

Several factors likely explain this superior performance:

High-quality, contract-level data: Unlike many previous studies relying on aggregated or survey data, this

research employs detailed administrative records from a real-life Tunisian insurer, ensuring data completeness

and behavioural accuracy.

Careful feature selection and data cleaning: Rigorous preprocessing and exclusion of irrelevant or noisy

variables improved model stability and reduced overfitting risk.

Balanced model complexity: Logistic regression provided a parsimonious yet powerful framework capable of

capturing the dominant behavioural and financial drivers of surrender while avoiding the instability and opacity

often observed in ensemble or neural models.

Contextual specificity: The Tunisian insurance market exhibits strong behavioural consistency and relatively

homogeneous product structures, which enhances the performance of parametric models like logistic regression.

This comparison also highlights a persistent research gap: while studies from Asia and Europe dominate the

literature, evidence from African insurance markets remains sparse and fragmented.

By delivering robust, contract-level results from Tunisia—an underexplored North African market—this study

not only achieves methodological excellence but also contributes new empirical insights to the broader

international discourse on surrender modelling.

RESULTS AND DISCUSSION

A. Logistic Regression Estimates

The logistic regression results reveal three major determinants of surrender behavior among Tunisian

policyholders. Prior partial surrender emerges as the most influential factor, increasing the likelihood of full

surrender by approximately six times (odds ratio ≈ 6.25). This finding strongly supports the behavioral

persistence hypothesis: policyholders who have previously accessed policy value through partial withdrawals

demonstrate substantially elevated risk of complete contract termination. This pattern may reflect ongoing

financial stress, reduced psychological commitment to the contract following initial withdrawal, or systematic

differences in financial planning sophistication between policyholders who utilize partial surrender options

versus those who do not.

The magnitude of this effect exceeds estimates from developed markets, where prior withdrawal behavior

typically increases surrender odds by factors of 2-3. This larger effect in Tunisia may reflect several contextual

factors. First, financial planning tools and alternative liquidity sources may be more limited in emerging markets,

making insurance contracts a primary emergency fund for middle-class households. Second, the financial

services regulatory environment in Tunisia underwent substantial evolution during the study period, potentially

creating uncertainty about contract features and encouraging precautionary behavior. Third, cultural factors

regarding savings and insurance may differ from Western contexts, affecting the psychological significance of

partial withdrawal decisions.

The existence of policy advances (loans) also increases surrender probability, with an odds ratio of

approximately 1.82. This finding aligns with the liquidity constraint hypothesis: policyholders utilizing policy

loans likely face financial pressure that elevates their overall propensity to terminate contracts. Policy loans

Page 3110

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

represent a halfway measure between full commitment (no cash access) and complete exit (surrender), and the

evidence suggests that policyholders exercising this option are substantially more likely to ultimately choose full

exit.

The specific magnitude of 1.82 implies that, holding other factors constant, contracts with outstanding policy

advances face 82% higher surrender odds than contracts without loans. This effect is economically meaningful

though smaller than the prior partial surrender effect. The distinction likely reflects the different implications of

these behaviors: partial surrenders permanently reduce policy value and cannot be reversed, while policy loans

can be repaid, maintaining the option value of the contract. Policyholders with loans but no history of partial

surrender may retain stronger attachment to their policies.

From a risk management perspective, the positive association between policy loans and surrender contradicts

the intuition that loans provide a 'release valve' reducing surrender pressure. Instead, the evidence suggests loans

serve as a leading indicator of financial distress rather than an effective retention mechanism. Insurers should

view active policy loan portfolios as early warning signals for elevated surrender risk rather than as successful

retention strategies.

Low mathematical reserves (below 80,000 Tunisian Dinars) are associated with significantly higher surrender

risk, with an odds ratio of approximately 0.56. The inverse odds ratio (< 1.0) indicates that higher reserve levels

are protective against surrender. This finding requires careful interpretation given the coding of the reserve

variable. Policies with reserves below the threshold face approximately 1/0.56 ≈ 1.79 times higher surrender

odds than policies with reserves above the threshold.

This inverse relationship between reserves and surrender risk aligns with multiple theoretical mechanisms. First,

reserves correlate strongly with policy duration—contracts held longer accumulate larger reserves through

premium payments and investment returns. Longer-duration policies face systematically lower surrender risk

due to sunk cost effects, increased attachment, and approach to maturity. Second, low reserves may signal poor

policy performance relative to expectations, reducing the opportunity cost of surrendering. Third, surrender

penalties are often structured as percentages of reserves, making early surrender particularly costly and deterring

exits for low-reserve policies.

The 80,000 TND threshold corresponds approximately to 3-5 years of premium payments for typical periodic-

savings products in the Tunisian market, based on prevailing contribution rates and interest crediting. This

suggests that the critical retention window extends well beyond the traditional 2-year persistency period

emphasized in agent compensation structures. Insurers should consider extending surrender penalties and

graduated bonus structures beyond the traditional early period to maintain retention incentives throughout the

contract lifecycle.

B. Model Performance Assessment

The logistic regression model demonstrates excellent discriminatory ability, achieving an AUC-ROC between

0.96 and 0.97 across multiple validation exercises. This performance substantially exceeds typical benchmarks

for logistic regression models in life insurance surrender prediction, which typically range from 0.70 to 0.85 in

the literature. Moreover, the achieved AUC rivals or exceeds advanced machine learning approaches including

random forests (typical AUC 0.85-0.92) and gradient boosted machines (typical AUC 0.88-0.95) reported in

recent studies.

Several factors likely contribute to this exceptional performance. First, the three selected predictors—prior

partial surrender, policy advances, and mathematical reserves—appear to capture the core behavioral and

financial dimensions of surrender risk with minimal noise. Each predictor represents a distinct theoretical

mechanism (behavioral persistence, liquidity stress, contract maturity/performance) while showing limited

multicollinearity. Second, the Tunisian market context may feature cleaner separation between surrendered and

persistent contracts than more mature insurance markets where surrender decisions reflect more complex,

multifaceted considerations.

Page 3111

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Third, data quality in the Tunisian insurance administrative systems may be particularly high due to recent

regulatory modernization efforts and information system investments. Complete, accurate data on policy

characteristics, transaction histories, and outcomes enable more precise modeling than datasets with substantial

missing information or measurement error. Fourth, the relatively concentrated market structure (23 domestic

companies) may create more homogeneous underwriting and product design practices, reducing unobserved

heterogeneity that would lower model performance in more fragmented markets.

Classification accuracy at optimal threshold selection exceeds 90%, with both high sensitivity (ability to identify

actual surrenders) and high specificity (ability to identify persistent contracts). This balanced performance across

both classes is noteworthy, as many real-world classification models achieve high overall accuracy primarily by

correctly predicting the majority class while performing poorly on the minority (surrendered) class. The model's

strong performance on both metrics indicates genuine discriminatory power rather than sample imbalance

artifacts.

Calibration assessment through visual inspection of predicted probability distributions suggests reasonable

correspondence between predicted probabilities and observed surrender frequencies across deciles. Predicted

probabilities span nearly the full 0-1 range, indicating the model produces differentiated risk assessments rather

than compressing predictions toward central values. This calibration quality is particularly important for capital

modeling and cash flow projection applications where absolute probability levels matter, not just relative risk

rankings.

C. Interpretation and Practical Implications

The findings emphasize the financial and behavioral dimensions of surrender risk and align with international

actuarial experience while revealing some distinctive features of the Tunisian market. The dominant role of prior

partial surrender behavior highlights the importance of early-stage monitoring and intervention. Insurers should

implement systematic tracking of partial withdrawal activity, flagging contracts for proactive retention outreach

following any partial surrender event.

The policy loan effect suggests that loan programs, while often marketed as retention tools, may actually serve

as early warning indicators of elevated surrender risk. Rather than relying on loans to reduce surrenders, insurers

should view loan portfolio composition and growth as key risk metrics informing reserve adequacy, liquidity

planning, and capital allocation decisions. Regular stress testing of surrender assumptions should incorporate

loan portfolio dynamics as a forward-looking indicator.

The reserve level findings have important implications for product design and pricing. The negative relationship

between reserves and surrender risk creates a 'J-curve' in surrender propensity over the policy lifecycle: moderate

risk at inception, elevated risk in early years when reserves remain low despite sunk costs accumulating, and

declining risk as reserves grow and maturity approaches. Product designers should consider graduated surrender

penalty structures aligned with this risk profile, potentially reducing penalties more slowly in middle contract

years than traditional approaches suggest.

Liquidity risk management requires scenario analyses incorporating the identified determinants. A 10% increase

in loan utilization combined with elevated partial surrender activity could trigger cascade effects substantially

increasing full surrender rates. Insurers should establish comprehensive monitoring dashboards tracking these

leading indicators in real-time, with predefined escalation procedures when metrics exceed predetermined

thresholds. Asset-liability management strategies must incorporate surrender risk scenarios reflecting the

estimated sensitivities.

Pricing and reserving should explicitly recognize heterogeneity in surrender risk based on observable

characteristics. Contracts with policy loans and/or prior partial surrender history require higher surrender

reserves than traditional age-duration cells would imply. Risk-based capital calculations under emerging African

solvency frameworks should incorporate these findings to ensure adequate capital buffers for surrender risk.

Page 3112

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Retention strategies should be targeted based on predicted surrender probabilities. Rather than uniform retention

efforts across all policyholders, insurers can prioritize high-risk segments (prior partial surrender, active loans,

low reserves) for intensive interventions including personalized communication, flexible payment arrangements,

and product modification options. Machine learning algorithms could further refine targeting by identifying

additional subtle patterns, though the logistic model's strong performance suggests limited gains available from

more complex approaches in this context.

COMPARISON WITH INTERNATIONAL LITERATURE

A. Alignment with Developed Market Research

The present findings show substantial consistency with developed market evidence on surrender determinants,

despite significant differences in market structure, regulation, and economic development between Tunisia and

advanced economies. The positive association between policy loans and surrender risk aligns with U.S. and

European research documenting that loan utilization signals financial distress and predicts elevated lapse rates.

The mathematical reserve effect parallels duration effects universally documented in the international literature,

where longer-duration contracts with accumulated value face systematically lower surrender risk.

However, the magnitude of effects differs meaningfully from developed market estimates. The odds ratio of

approximately 6.25 for prior partial surrender substantially exceeds typical estimates of 2-3 in Western markets.

This larger effect may reflect limited financial sophistication and fewer alternative liquidity sources in emerging

markets, making the initial decision to access policy value a stronger signal of fundamental financial stress.

The interest rate sensitivity documented in developed markets is not directly tested in the present study due to

focus on contract-level characteristics rather than macroeconomic variables. However, the stability of findings

across the 2010-2024 period—which encompasses both declining rates (2010-2020) and rising rates (2021-

2024)—suggests the identified microeconomic factors operate relatively independently of short-term interest

rate movements. This contrasts with developed market evidence showing strong surrender rate sensitivity to rate

changes, possibly reflecting less developed alternative investment markets in Tunisia reducing opportunity cost

effects.

B. Emerging Market Specificities

The comparison with the Nepal study by Ghimire et al. (2024) reveals both commonalities and differences across

emerging markets. Both studies identify economic factors (liquidity needs) as primary drivers, with knowledge

gaps and behavioral factors playing supporting roles. However, the Tunisian analysis finds stronger effects for

observable financial indicators (reserves, loans, prior surrenders) than the Nepal survey research emphasizes.

This difference likely reflects methodological distinctions: administrative data analysis captures revealed

preferences through actual financial behaviors, while survey responses may emphasize socially acceptable

explanations (advice from family, lack of information) over potentially embarrassing financial difficulties.

The finding that demographic variables show weak associations with surrender aligns with Ghimire et al.'s Nepal

results but contrasts with some developed market studies emphasizing age, gender, and education effects. This

convergence across diverse emerging markets (South Asia, North Africa) suggests that in developing economy

contexts, financial and behavioral factors dominate over sociodemographic characteristics in predicting

surrender. This pattern may reflect higher within-demographic-group variance in financial circumstances and

insurance understanding in emerging markets compared to more economically homogeneous developed markets.

C. Methodological Performance Comparison

The achieved AUC of 0.96-0.97 positions this study at the upper end of the performance distribution in the

surrender prediction literature. This performance exceeds most reported logistic regression results (typical AUC

0.70-0.85) and rivals advanced machine learning approaches. The strong performance of the parsimonious three-

predictor logistic model in Tunisia suggests that in some contexts, careful variable selection based on theoretical

considerations may achieve performance rivaling data-driven machine learning approaches using many more

Page 3113

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

features. This finding has practical importance for insurers with limited data science resources, validating the

continued relevance of traditional actuarial methods when properly implemented.

However, the exceptionally high AUC also warrants cautious interpretation. The clean separation between

surrendered and persistent contracts may reflect specific features of the Tunisian periodic-savings market that

would not generalize to other products (unit-linked, variable annuities) or markets with more heterogeneous

policyholder populations. External validation using data from other African markets would strengthen

confidence in the generalizability of findings.

PROFESSIONAL RECOMMENDATIONS

A. Early Warning Systems and Monitoring

Insurers should develop comprehensive surrender risk monitoring systems incorporating the identified leading

indicators:

 Real-time dashboards tracking partial surrender frequency, policy loan utilization rates, and reserve

distribution across the portfolio. Establish quantitative thresholds triggering management review and

proactive intervention protocols.

 Individual contract scoring models assigning surrender risk scores to each policy based on the logistic

regression framework. Policies scoring above predetermined thresholds enter intensive retention

programs with enhanced customer service, personalized communication, and targeted retention offers.

 Cohort analysis tracking surrender rates across policy vintages, distribution channels, and product

variations to identify systematic patterns requiring product design or underwriting adjustments.

B. Proactive Retention Strategies

Targeted retention programs should focus on high-risk segments:

 Post-partial-surrender intervention: Following any partial surrender event, customer service

representatives should proactively contact policyholders to understand circumstances, provide education

on policy features, and explore alternatives to full surrender including premium payment holidays, sum

assured reductions, or conversion to paid-up status.

 Loan portfolio management: Implement systematic loan repayment reminders, offer flexible repayment

schedules, and provide education on loan implications for death benefits and surrender values. Consider

restrictions on combined loans and partial surrenders to limit cumulative erosion of policy value.

 Low-reserve policy support: For contracts in early years with low reserves, consider special retention

bonuses, loyalty guarantees, or premium subsidies to increase policyholder engagement and reduce

financial strain during vulnerable periods.

C. Product Design and Pricing Innovations

 Graduated surrender penalty structures aligned with the identified risk profile, maintaining meaningful

penalties through middle contract years when reserve growth creates increasing surrender exposure.

 Partial surrender limitations restricting frequency and magnitude of withdrawals to mitigate behavioral

cascade effects. Consider annual limits, cumulative lifetime caps, or increasing restrictions following

initial withdrawals.

 Embedded retention incentives including terminal bonuses, duration-linked guarantees, and loyalty

benefits that increase opportunity cost of early surrender. Market value adjustment clauses that adjust

surrender values for interest rate changes may reduce rate-driven surrenders.

Page 3114

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

 Risk-based pricing incorporating surrender risk factors explicitly in premium calculations rather than

assuming uniform persistency across contracts. Policies sold with loan provisions or flexible withdrawal

features should carry higher premiums reflecting increased embedded option value.

D. Liquidity and Capital Management

 Stress testing frameworks incorporating correlated scenarios combining increased loan utilization,

elevated partial surrender rates, and rising full surrender frequency. Quantify potential liquidity needs

under adverse scenarios and maintain appropriate buffers.

 Dynamic rebalancing strategies adjusting portfolio liquidity based on leading indicators. When partial

surrender rates or loan utilization increase significantly, shift asset allocation toward more liquid

securities to facilitate potential redemption needs.

 Solvency capital allocation under IFRS 17 and emerging African solvency frameworks should reflect the

heterogeneous risk profile identified through the logistic model. Risk-based capital requirements for

surrender risk should vary based on portfolio composition across the identified risk dimensions.

E. Customer Communication and Financial Literacy

 Comprehensive policy education at inception and periodically thereafter, explaining surrender penalties,

policy loan mechanics, partial surrender implications, and opportunity costs of early termination. Use

multiple communication channels including digital platforms, mobile apps, and in-person consultations.

 Financial planning integration positioning life insurance within broader household financial management

rather than as standalone products. Provide tools and resources helping policyholders understand

insurance's role in emergency funds, retirement planning, and estate planning.

 Behavioral nudges leveraging insights from behavioral economics to encourage persistence, such as loss-

framing communications emphasizing costs of surrendering, commitment devices making surrenders

more psychologically difficult, or social proof messaging highlighting persistence norms.

LIMITATIONS AND FUTURE RESEARCH DIRECTIONS

A. Study Limitations

This study is subject to several important limitations that warrant acknowledgment and suggest directions for

future research:

 Geographic and temporal scope: The analysis focuses exclusively on the Tunisian life insurance market

between 2010 and 2024. While Tunisia represents a significant emerging market, generalization to other

African countries, Middle Eastern markets, or other emerging economies requires caution given

institutional, regulatory, and cultural differences. The 14-year observation period encompasses

substantial macroeconomic volatility including the Arab Spring political transition, but may not capture

full economic cycles.

 Product focus: The exclusive focus on periodic-savings products limits applicability to other life

insurance categories including pure term insurance, unit-linked products, and variable annuities.

Surrender dynamics likely differ substantially across product types due to variations in investment risk

allocation, surrender penalty structures, and target market characteristics.

 Variable limitations: The analysis omits several potentially important determinants due to data

availability constraints. Macroeconomic variables including interest rates, inflation, unemployment, and

GDP growth were not incorporated despite extensive evidence from developed markets documenting

Page 3115

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

their importance. Future research should integrate contract-level microeconomic factors with time-

varying macroeconomic conditions to assess relative importance and interaction effects.

 Policyholder characteristics including income, education, occupation quality, marital status, and

household composition were unavailable or incompletely recorded in administrative systems. These

sociodemographic variables may reveal additional surrender predictors or moderate the effects of

financial variables. Integrating administrative data with survey research or external demographic

databases would enrich understanding.

 Behavioral and attitudinal measures captured through surveys in other studies—including financial

literacy, trust in insurance companies, satisfaction with agent service, and knowledge of policy features—

are absent from administrative data. Mixed-methods research combining quantitative contract data with

qualitative insights would provide more comprehensive understanding of surrender motivations.

 Methodological limitations: The binary logistic regression approach, while interpretable and achieving

strong performance, may not capture non-linear relationships or complex interactions between predictors

as flexibly as machine learning algorithms. The three-predictor specification prioritizes parsimony and

practical implementability over exhaustive modeling of all potential effects.

 Selection and endogeneity concerns: The observational study design precludes strong causal inference.

Prior partial surrender and policy loan utilization are endogenous to underlying financial circumstances

and preferences that also drive surrender decisions, potentially biasing estimated effects. Instrumental

variable approaches or natural experiments would strengthen causal identification.

B. Future Research Directions

Several promising avenues for extending this research emerge:

 Multi-country comparisons: Replicating the analysis across multiple African and Middle Eastern

insurance markets would assess generalizability and identify market-specific versus universal surrender

determinants. Systematic cross-country comparisons could illuminate the role of regulatory frameworks,

distribution systems, and cultural factors in shaping surrender behavior.

 Longitudinal survival analysis: Rather than binary surrender/persistence outcomes, time-to-surrender

analysis using Cox proportional hazards or parametric survival models would reveal how risk evolves

over policy duration and respond to time-varying covariates. Understanding the temporal dynamics of

surrender risk enables more sophisticated ALM and capital modeling.

 Machine learning extensions: While the logistic model performs strongly, exploring ensemble methods,

neural networks, and causal machine learning approaches may uncover subtle patterns or improve

probability calibration. Explainable AI techniques (SHAP values, LIME) could maintain interpretability

while leveraging advanced algorithms.

 Behavioral experiments: Randomized controlled trials testing alternative communication strategies,

retention incentives, or product design features would provide credible causal evidence on intervention

effectiveness. Field experiments embedded within insurer operations offer opportunities for rigorous

evaluation of retention programs.

 Macro-micro integration: Hierarchical models incorporating both contract-level characteristics and time-

varying macroeconomic conditions would assess relative importance and potential interactions. Does the

prior partial surrender effect strengthen or weaken during periods of rising interest rates or economic

stress?

Page 3116

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

 Life event tracking: Integrating administrative insurance data with external sources capturing major life

events (marriage, divorce, childbirth, unemployment, health events) would test the emergency fund

hypothesis more directly and enable more proactive risk-based interventions.

CONCLUSION

This study provides the first comprehensive contract-level analysis of life insurance surrender determinants in

Tunisia, contributing important empirical evidence from an understudied emerging African market. Using a

parsimonious three-predictor logistic regression model, the analysis identifies prior partial surrender, policy

advance existence, and mathematical reserve levels as the primary drivers of surrender decisions among Tunisian

periodic-savings policyholders. The model achieves exceptional discriminatory performance (AUC 0.96-0.97),

rivaling or exceeding advanced machine learning approaches reported in developed market studies.

The findings demonstrate substantial consistency with international actuarial research while revealing some

distinctive features of emerging market contexts. The particularly strong effect of prior partial surrender behavior

suggests that initial liquidity access events have profound implications for subsequent contract persistence in

developing economies with limited alternative financial instruments. The positive association between policy

loans and surrenders challenges conventional wisdom that loan provisions effectively reduce surrender risk,

instead suggesting they serve primarily as early warning indicators of financial distress.

From a practical perspective, the results provide actionable intelligence for insurers seeking to manage surrender

risk more effectively. Early warning systems monitoring partial surrender activity, policy loan utilization, and

reserve accumulation enable proactive retention interventions targeting high-risk contracts. Product design

innovations including graduated surrender penalties, partial withdrawal restrictions, and embedded loyalty

incentives can reduce surrender propensity while maintaining policyholder value propositions. Risk-based

pricing and capital allocation frameworks should explicitly incorporate the identified heterogeneity in surrender

risk rather than assuming uniform persistency.

The study contributes to the limited but growing literature on emerging market insurance by applying rigorous

quantitative methods to comprehensive administrative data from North Africa. The findings suggest that while

fundamental economic and behavioral drivers of surrender operate similarly across diverse contexts, the

magnitude of effects and relative importance of specific factors vary meaningfully between developed and

emerging markets. This heterogeneity has important implications for international insurers expanding into

African markets, who cannot simply transplant developed market risk models but must calibrate approaches to

local contexts.

Methodologically, the strong performance of a simple logistic model with carefully selected theoretically-

motivated predictors challenges the notion that surrender prediction necessarily requires complex machine

learning algorithms. While advanced methods offer advantages in some contexts, the results validate the

continued relevance of traditional actuarial approaches when thoughtfully implemented. This finding has

practical importance for insurers with limited data science resources, particularly in emerging markets where

technical capacity constraints may limit adoption of sophisticated analytical methods.

Looking forward, insurance market development in Tunisia and across Africa requires improved understanding

of policyholder behavior, product preferences, and risk dynamics. The present research provides a

methodological template and empirical benchmark for future studies while highlighting the need for multi-

country comparisons, macro-micro integration, and mixed-methods approaches combining quantitative

administrative data with qualitative behavioral insights. As African insurance markets continue growing and

regulatory frameworks evolve, evidence-based risk management grounded in local data will become increasingly

essential for sustainable industry development.

Page 3117

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

REFERENCES

1. Albizzati, M. O., & Geman, H. (1994). Interest rate risk management and valuation of the surrender

option in life insurance policies. Journal of Risk and Insurance, 61(4), 616-637.

2. Azzone, M., Barucci, E., Giuffra Moncayo, G., & Marazzina, D. (2022). A machine learning model for

lapse prediction in life insurance contracts. Expert Systems with Applications, 191, 116261.

3. Bacinello, A. R., Biffis, E., Millossovich, P., Olivieri, A., & Pitacco, E. (2017). Intensity-based

frameworks for surrender modeling in life insurance. Insurance: Mathematics and Economics, 78, 212-

227.

4. Becker, B., Opp, M. M., & Saidi, F. (2022). Regulatory forbearance in the US insurance industry: The

effects of eliminating capital requirements. Review of Financial Studies, 35(12), 5438-5482.

5. Bobtcheff, C., Casamatta, C., & Villamil, A. P. (2016). Financial intermediation and the regulation of

life insurance companies. Journal of Financial Intermediation, 28, 118-132.

6. Boonmeekham, A., Phoonsawat, N., & Klongdee, W. (2019). Logistic regression model for lapse

analysis of life insurance policy. Burapha Science Journal, 24(2), 730-741.

7. Breiman, L. (1984). Classification and Regression Trees. Wadsworth International Group.

8. Briys, E., & de Varenne, F. (1997). On the risk of insurance liabilities: Debunking some common pitfalls.

Journal of Risk and Insurance, 64(4), 673-694.

9. Chang, E. C., & Schmeiser, H. (2021). Valuing the option to surrender in life insurance. Journal of

Banking & Finance, 128, 106142.

10. Cheng, C., Ding, Y., & Zhou, C. (2023). Surrender contagion in life insurance. European Journal of

Operational Research, 304(3), 1234-1248.

11. Cox, S. H., Lin, Y., & Pedersen, H. (1992). Mortality tables and lapse rates. Transactions of Society of

Actuaries, 44, 489-569.

12. Dar, A., & Dodds, C. (1989). Interest rates, the emergency fund hypothesis and saving through

endowment policies: Some empirical evidence for the UK. Journal of Risk and Insurance, 56(3), 415-

433.

13. D'Ortona, N. E., & Staffa, M. S. (2016). The theoretical surrender value in life insurance. Insurance

Markets and Companies, 7(1), 31-44.

14. EIOPA (European Insurance and Occupational Pensions Authority). (2019). Financial Stability Report.

European Union.

15. Eling, M., & Kiesenbauer, D. (2014). What policy features determine life insurance lapse? An analysis

of the German market. Journal of Risk and Insurance, 81(2), 241-269.

16. Eling, M., & Kochanski, M. (2013). Research on lapse in life insurance: What has been done and what

needs to be done? Journal of Risk Finance, 14(4), 392-413.

17. European Central Bank. (2009). Financial Stability Review. Frankfurt: European Central Bank.

18. European Central Bank. (2017). Financial Stability Review, May 2017. Frankfurt: European Central

Bank.

19. Fang, H., & Kung, E. (2012). Why do life insurance policyholders lapse? The roles of income, health

and bequest motive shocks. NBER Working Paper No. 17899, National Bureau of Economic Research.

20. Farkas, S., Fringuellotti, F., & Tunaru, R. (2023b). Life insurance surrender risk. Journal of Banking &

Finance, 147, 106729.

21. Foley-Fisher, N., Heinrich, N., & Verani, S. (2022). How do U.S. life insurers manage liquidity in times

of stress? FEDS Notes, Board of Governors of the Federal Reserve System.

22. The Geneva Association. (2012). Surrenders in the Life Insurance Industry and their Impact on Liquidity.

Geneva: The International Association for the Study of Insurance Economics.

23. Ghimire, R., Thapa, B. S., & Waikar, V. (2024). Policyholder surrender behavior in the Nepalese life

insurance market during the COVID-19 pandemic. Interdisciplinary Journal of Innovation in Nepalese

Academia, 3(2), 86-105.

24. Gemmo, I., Götz, M., & Schmeiser, H. (2018). Life insurance and demographic change: An empirical

analysis of surrender decisions based on panel data. SSRN Working Paper No. 3230274.

25. International Monetary Fund. (2021). Global Financial Stability Report. Washington, DC: IMF.

26. International Monetary Fund. (2024). Implementing Risk-Based Solvency for Insurers—Lessons from

Kenya, Mexico, and South Africa. Washington, DC: IMF.

Page 3118

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

27. Jansen, M. (2021). Solvency risk and policyholder behavior in the life insurance sector. Journal of

Financial Stability, 54, 100876.

28. Kgare, M. (2021). Predicting Lapse Rate in Life Insurance Using Machine Learning Algorithms. Master's

thesis, University of South Africa.

29. Kiesenbauer, D. (2012). Main determinants of lapse in the German life insurance industry. North

American Actuarial Journal, 16(1), 52-73.

30. Kiermayer, M., Krah, F., Pfeuffer, M., & Weiß, A. (2021). Modeling surrender risk in life insurance:

Theoretical and experimental insight. arXiv preprint arXiv:2101.11590.

31. Koijen, R. S., Van Nieuwerburgh, S., & Yogo, M. (2022). Why are life insurers exposed to interest rate

risk? Review of Financial Studies, 35(12), 5374-5437.

32. Kubitza, C., Grochola, N., & Grill, M. (2023). Life insurance convexity. ECB Working Paper No. 2829,

European Central Bank.

33. Kuo, W., Tsai, C., & Chen, W. K. (2003). An empirical study on the lapse rate: The cointegration

approach. Journal of Risk and Insurance, 70(3), 489-508.

34. Leiria, M., Dionísio, A., & Ferreira, P. (2021). Non-life insurance cancellation: A systematic quantitative

literature review. The Geneva Papers on Risk and Insurance-Issues and Practice, 46(4), 882-914.

35. Loisel, S., Piette, P., & Tsai, J. C. H. (2019). Applying economic measures to lapse risk management

with machine learning approaches. ASTIN Bulletin, 49(3), 839-871.

36. Manteigas, C., Neves, C., & Gonçalves, L. (2024). Understanding and predicting lapses in mortgage life

insurance: A data-driven approach. Expert Systems with Applications, 236, 121346.

37. National Association of Insurance Commissioners (NAIC). (2021). Liquidity Risk Management

Guidance. Kansas City: NAIC.

38. Ozdagli, A., & Wang, Z. (2019). Interest rates and insurance company investment behavior. Harvard

Business School Working Paper.

39. Pitacco, E. (2010). Teaching life insurance mathematics: From the lx's to the risk management approach.

ICA 2010 Education/Professionalism Paper.

40. Poufinas, T., & Michaelide, G. (2018). Determinants of life insurance policy surrenders. Modern

Economy, 9, 1400-1422.

41. Russell, D. T., Fier, S. G., Carson, J. M., & Dumm, R. E. (2013). An empirical analysis of life insurance

policy surrender activity. Journal of Insurance Issues, 36(1), 35-57.

42. Schott, F. H. (1971). Disintermediation through policy loans at life insurance companies. Journal of

Finance, 26(3), 719-729.

43. Shamsuddin, S. N., Mohmad, S., Hanafi, N. H., Samian, M. H., Juanis, D. J., & Zahir, I. N. (2025).

Lapsation logistic regression model: A case in life insurance. International Journal of Research and

Innovation in Social Science, 9(7), 4179-4190.

44. Swiss Re Institute. (2023). Sigma Report: Global Insurance Review. Zurich: Swiss Re.

45. Tsai, C., Kuo, W., & Chen, W. K. (2002). Early surrender and the distribution of policy reserves.

Insurance: Mathematics and Economics, 31(3), 429-445.

46. Wooldridge, J. M. (2002). Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT

Press.

47. Zelinová, S., et al. (2024). The impact of dynamic surrender on guarantees and profitability of life

insurance. TEM Journal, 13(3), 2502-2511.

Page 3119