{"paper_id":"2609.39569","title":"Spouse-Protected Tontines: Household Decumulation via Neural-Network Optimization","authors":["Duy-Minh Dang","Yukan Perumal"],"paper_date":"2026-09-30","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2609.39569","page_url":"https://thequant.space/flowcharts/spouse-protected-tontines-household-decumulation-via-neural-network-optimization/","scores":{"empirical_rigor":7,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a spouse‑protected tontine contract that remains active until household extinction, with a neural‑network based global‑in‑time policy for withdrawals and asset allocation, and derives actuarial‑fairness and mortality‑credit allocation rules.","quote":"We develop a spouse-protected tontine in which a first death changes the household state but generates no pool transfer. We develop a numerical solution method for this problem based on a global-in-time neural-network parameterization of admissible control policies.","summary":"We develop a spouse-protected tontine in which a first death changes the household state but generates no pool transfer. The same account remains attached to the household contract until extinction and funds a spouse-only continuation phase if the retiree dies first. We derive contract-level…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"A homogeneous large‑pool approximation is used to obtain a representative household decumulation problem.","type":"model","stated":"explicit","quote":"Under a homogeneous large-pool approximation, we formulate a multidimensional decumulation problem for a representative household, with withdrawal and rebalancing controls while the retiree is alive.","support":"full"},{"assumption":"Annual decision times are assumed for withdrawals, rebalancing, and mortality‑credit allocation.","type":"execution","stated":"explicit","quote":"We consider the finite planning horizon [0, T] with annual decision times T = {tm | tm = m, m = 0, . . . , M}, M = T, at which mortality credits are allocated, fees are deducted, household payments are made, and portfolio positions are rebalanced.1 Here, t0 = 0 is the inception date.","support":"full"},{"assumption":"The contract wealth may become non‑positive and is interpreted as accumulated debt under an insolvency convention.","type":"model","stated":"explicit","quote":"The quantity Wt is allowed to become non-positive; under the insolvency convention, negative wealth is interpreted as accumulated debt.","support":"full"},{"assumption":"Mortality‑credit gain rates depend on the household state (both alive, retiree‑only, spouse‑only).","type":"model","stated":"explicit","quote":"Under this design, pool-exit probabilities and mortality-credit gain rates are household-state dependent.","support":"full"},{"assumption":"The objective balances expected cumulative real household payments against CVaR of terminal shortfalls without conditioning on survival to the horizon.","type":"statistical","stated":"explicit","quote":"The objective balances expected cumulative real household payments against the Conditional Value-at-Risk (CVaR) of terminal shortfalls relative to household reserve targets, without conditioning on survival to the horizon. Both criteria are evaluated without conditioning on survival to the horizon.","support":"full"},{"assumption":"State‑contingent terminal reserve targets are exogenously specified and used to define shortfall.","type":"model","stated":"explicit","quote":"Specifically, let w∗ 11, w∗ 10, and w∗ 01 be nonnegative terminal reserve targets, corresponding respectively to the terminal household states 11, 10, and 01. These reserve targets are model inputs and may be specified externally using household retirement-spending benchmarks.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Four‑asset monthly return panel (Australian equities, Australian government bonds, U.S. eq","provider":null,"kind":"market-prices","access":"public","period":"1935‑01 to 2022‑12","universe":"Australian and U.S. asset classes","quote":"We use a four-asset monthly return panel comprising Australian equities, Australian government bonds, U.S. equities converted to AUD, and U.S. 30-day Treasury bills converted to AUD. Following [28], the four series are aligned over the common sample period 1935:1–2022:12 and converted to real AUD returns using Australian CPI.","support":"full"},{"name":"2021 Australian male and female period life tables (Human Mortality Database)","provider":"Human Mortality Database","kind":"macro","access":"public","period":"2021","universe":"Australian population","quote":"For deterministic mortality, we use the 2021 single-year Australian male and female period life tables from the Human Mortality Database [21].","support":"full"}],"known":[{"id":"bloomberg","name":"Bloomberg","provider":"Bloomberg L.P.","kind":"market-prices","access":"commercial","quote":"The Australian equity series is a long-horizon capitalization-weighted total-return index constructed from Bloomberg All Ordinaries data together with historical dividend-yield information from [25] and the Reserve Bank of Australia.","mentions":2,"used_in":2}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"A yearly tontine management fee ϱ∈(0,1) is applied proportionally to the account balance each year.","quote":"(3.5) We allow for a baseline tontine management fee, deducted once per year at times tm, m = 1, . . . , M, and modeled as a proportional charge on the account balance. Let ϱ ∈(0, 1) denote the yearly fee rate, and define ϱm = ϱ 1{ m≥1, f Wm−>0, Hm−̸=00 }, m = 0, . . . , M. (3.6) The contract wealth after fee deduction is Wm−= (1 −ϱm) f Wm−, f Wm−given in (3.5).","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"A yearly tontine management fee ϱ∈(0,1) is applied proportionally to the account balance each year.","card_quote":"(3.5) We allow for a baseline tontine management fee, deducted once per year at times tm, m = 1, . . . , M, and modeled as a proportional charge on the account balance. Let ϱ ∈(0, 1) denote the yearly fee rate, and define ϱm = ϱ 1{ m≥1, f Wm−>0, Hm−̸=00 }, m = 0, . . . , M. (3.6) The contract wealth after fee deduction is Wm−= (1 −ϱm) f Wm−, f Wm−given in (3.5).","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"none","findings":[{"finding":"Spouse‑only continuation is a first‑order and persistent household event in the calibrated Australian experiments.","kind":"theoretical","quote":"Numerical experiments calibrated to Australia show that the spouse-only phase is a first-order and persistent household event.","support":"full"},{"finding":"Book‑level diversification removes most of the representative‑contract excess upper‑tail cost of spouse continuation (≈97% reduction) while leaving expected per‑contract continuation cost unchanged.","kind":"performance","quote":"In these experiments, book-level diversification removes most of the representative-contract excess upper-tail cost of spouse continuation in the large-book limit, without changing expected per-contract continuation cost. In the baseline capped experiments, book-level diversification reduces the representative-contract excess upper-tail cost by approximately 97% in the conditional large-book limit across the","support":"full"},{"finding":"The expected per‑contract continuation cost varies little across the selected learned policies under capped continuation.","kind":"performance","quote":"The numerical results show that spouse continuation is frequent and persistent, with a substantial expected cost that varies little across the selected learned policies under capped continuation.","support":"full"},{"finding":"Cross‑country mortality comparison shows the prevalence and persistence of the spouse‑only phase are not specific to Australia.","kind":"theoretical","quote":"A cross-country mortality comparison indicates that the prevalence and persistence of the spouse-only phase are not specific to Australia. The cross-country mortality comparison indicates that the prevalence and persistence of the spouse-only phase are not specific to Australia.","support":"full"},{"finding":"Global‑in‑time neural‑network policies successfully approximate the solution to the multidimensional reward‑risk control problem.","kind":"theoretical","quote":"Admissible global-in-time neural-network policies approximate the solution of the resulting multidimensional reward–risk problem.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":16,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":2,"extracted_at":"2026-11-14T19:39:24Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}