ArticlePublisher preview available

Recovery process optimization using survival regression

Authors:
Jiří Witzany
Jiří Witzany
Jiří Witzany
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Anastasiia Kozina
Anastasiia Kozina
Anastasiia Kozina
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Read publisher preview
Download citation
To read the full-text of this research, you can request a copy directly from the authors.

Abstract and Figures

The goal of this paper is to propose, empirically test and compare different logistic and survival analysis techniques in order to optimize the debt collection process. This process uses various actions, such as phone calls, mails, visits, or legal steps to recover past due loans. We focus on the soft collection part, where the question is whether and when to call a past-due debtor with regards to the expected financial return of such an action. We propose to use the survival analysis technique, in which the phone call can be compared to a medical treatment, and repayment to the recovery of a patient. We show on a real banking dataset that, unlike ordinary logistic regression, this model provides the expected results and can be efficiently used to optimize the soft collection process.
Figures - available from: Operational Research
This content is subject to copyright. Terms and conditions apply.
A preview of this full-text is provided by Springer Nature.
Learn more
Content available from Operational Research
This content is subject to copyright. Terms and conditions apply.
Vol.:(0123456789)
Operational Research (2022) 22:5269–5296
https://doi.org/10.1007/s12351-022-00703-3
1 3
ORIGINAL PAPER
Recovery process optimization using survival regression
JiříWitzany1 · AnastasiiaKozina1
Received: 14 December 2020 / Revised: 31 January 2022 / Accepted: 6 March 2022 /
Published online: 29 March 2022
© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022
Abstract
The goal of this paper is to propose, empirically test and compare different logistic
and survival analysis techniques in order to optimize the debt collection process.
This process uses various actions, such as phone calls, mails, visits, or legal steps
to recover past due loans. We focus on the soft collection part, where the question
is whether and when to call a past-due debtor with regards to the expected financial
return of such an action. We propose to use the survival analysis technique, in which
the phone call can be compared to a medical treatment, and repayment to the recov-
ery of a patient. We show on a real banking dataset that, unlike ordinary logistic
regression, this model provides the expected results and can be efficiently used to
optimize the soft collection process.
Keywords Decision support systems· Credit risk modeling· Survival analysis·
Scoring· Debt recovery
JEL Classification G21· G28· C14
1 Introduction
The recovery process has become an important part of the banking business model.
Its main task is to manage overdue receivables through various enforcement tools,
with the goal of maximizing the final recovery. At present, due to growing portfolios
and in order to streamline all the activities performed, in particular those related to
the retail segments, banks are trying to make most of the daily recurring processes
as automated and efficient as possible. The recovery process is, in this respect, no
exception, and, therefore, modifications and improvements are constantly being
developed.
* Jiří Witzany
jiri.witzany@vse.cz
1 Faculty ofFinance andAccounting, Prague University ofBusiness andEconomics, W.
Churchill Sq. 4, 130 67, Prague, CzechRepublic
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
ResearchGate has not been able to resolve any citations for this publication.
A Markov decision model for consumer term-loan collections
Article
Full-text available
  • May 2019
  • Rev Quant Finance Account
We examine how to efficiently schedule collection actions for consumer term-loan accounts over time using a Markov decision model. A consumer loan account at each age can be classified into different account states, including current, delinquent, early payoff, default, and bankrupt. We model state transitions of loan accounts using a Markov transition matrix, and develop an optimization method to determine the collection action at each state and age for each consumer type to maximize the lender’s expected value. The optimization approach incorporates default risk and operational cost, and also addresses the time value of money, the tradeoff between interest revenue and borrowing cost, time consistency in optimization, competing risks between different account states, and penalty for late payment. Compared with a static collection policy, our method is demonstrably more valuable for accounts with high interest rates and medium to high loan amount, especially with stronger collection effects. We also demonstrate how the collection actions implemented under an optimal collection policy are affected by interest rate, loan amount, and collection effects.
View
View
View
Dynamic Credit-Collections Optimization
Article
  • Mar 2019
Based on a dynamic model of the stochastic repayment behavior exhibited by delinquent credit-card accounts in the form of a self-exciting point process, a bank can control the arrival intensity of repayments using costly account-treatment actions. A semi-analytic solution to the corresponding stochastic optimal control problem is obtained using a recursive approach. For a linear cost of treatment effort, the optimal policy in the two-dimensional (intensity, balance) space is described by the frontier of a convex action region. The unique optimal policy significantly reduces a bank’s loss given default and concentrates the collection effort onto the best possible actions at the best possible times so as to minimize the sum of the expected discounted outstanding balance and the discounted cost of the collection effort, thus maximizing the net value of any given delinquent credit-card account. This paper was accepted by Noah Gans, stochastic models and simulation.
View
Identifying hidden patterns in credit risk survival data using Generalised Additive Models
Article
  • Mar 2019
  • EUR J OPER RES
Modelling patterns in credit risk using survival analysis techniques have received considerable and increasing attention over the past decade. In these models, the predictor of the hazard of default is often expressed as a simple linear combination of the risk factors. In this work, we discuss how these models can be enhanced using Generalised Additive Models (GAMs). In the GAMs framework, the predictor is formulated as a combination of flexible univariate functions of the risk factors. In this paper, we parametrise GAMs for credit risk data in terms of penalised splines, outline the implementation via frequentist and Bayesian MCMC methods, apply them to a large portfolio of credit card accounts, and show how GAMs can be used to improve not only the application, behavioural and macro-economic components of survival models for credit risk data at individual account level, but also the accuracy of predictions. From a practitioner point of view, this work highlights that some accounts may actually become more (less) attractive to the lender if flexible smooth functions are used whereas the same applicant may be denied (accepted) a loan if the linearity assumption is forced.
View
Dynamic survival models with varying coefficients for credit risks.
Article
  • Nov 2018
  • EUR J OPER RES
Single event survival models predict the probability that an event will occur in the next period of time, given that the event has not happened before. In the context of credit risk, where one may wish to predict the probability of default on a loan account, such models have advantages over cross sectional models. The literature shows that the parameters of such models changed after compared with before the financial crisis of 2008. But there is also the possibility that the sensitivity of the probability of default, to say behavioural variables, changes over the life of an account. In this paper we make two contributions. First, we parameterise discrete time survival models of credit card default using B-splines to represent the baseline relationship. These allow a far more flexible specification of the baseline hazard than has been adopted in the literature to date. This baseline relationship is crucial in discrete time survival models and typically has to be specified ex-ante. Second, we allow the estimates of the parameters of the hazard function to themselves be a function of duration time. This allows the relationship between covariates and the hazard to change over time. Using a large sample of credit card accounts we find that these specifications enhance the predictive accuracy of hazard models over specifications which adopt the type of baseline specification in the current literature and which assume constant parameters.
View
Data-Driven Consumer Debt Collection via Machine Learning and Approximate Dynamic Programming
Article
  • Jan 2018
This paper develops and tests a framework for the data-driven scheduling of outbound calls made by debt collectors. These phone calls are used to persuade debtors to settle their debt, or to negotiate payment arrangements in case debtors are willing, but unable to repay. We determine on a daily basis which debtors should be called to maximize the amount of delinquent debt recovered in the long term, under the constraint that only a limited number of phone calls can be made each day. Our approach is to formulate a Markov decision process and, given its intractability, approximate the value function based on historical data through the use of state-of-the-art machine learning techniques. Specifically, we predict the likelihood with which a debtor in a particular state is going to settle its debt and use this as a proxy for the value function. Based on this value function approximation, we compute for each debtor the marginal value of making a call. This leads to a particularly straightforward optimization procedure, namely, we prioritize the debtors that have the highest marginal value per phone call. We validate our proposed methodology in a controlled field experiment conducted with real debtors. The results show that our optimized policy substantially outperforms the current scheduling policy that has been used in business practice for many years. Most importantly, our policy collects more debt in less time, whilst using substantially fewer resources — leading to a large increase in the amount of debt collected per phone call.
View
Debtor level collection operations using Bayesian dynamic programming
Article
  • Jun 2018
  • J OPER RES SOC
After a borrower defaults on their repayment obligations, collectors of unsecured consumer credit debt have a number of actions they can take to secure some repayment of the debt. The operations management challenge in this setting is to decide which of these actions to take, how long to take them, and in what sequence to take them, in order to maximize the recovery rate. In this paper, we adopt a dynamic programming approach to find an optimal policy of which action to undertake in the next period, using Bayesian updating to take into account the individual debtor’s repayment performance thus far. The use of the model is empirically illustrated using data provided by a European bank’s in-house collections department.
View
View
View