{"paper_id":"2306.08809","title":"Optimal Portfolio Execution in a Regime-switching Market with Non-linear Impact Costs: Combining Dynamic Program and Neural Network","authors":["Xiaoyue Li","John M. Mulvey"],"paper_date":"2023-06-15","paper_type":"methodological","paper_url":"https://arxiv.org/pdf/2306.08809v1","page_url":"https://thequant.space/flowcharts/optimal-portfolio-execution-in-a-regime-switching-market-wit/","scores":{"empirical_rigor":3.5,"math_complexity":8,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The paper proposes a combined numerical method that first solves approximated orthogonal portfolios with dynamic programming and then refines the solution using a neural network for optimal multi‑asset execution in a regime‑switching market.","quote":"First, we calculate the approximated orthogonal portfolios based on estimated impact cost functions; second, we employ dynamic program to learn the optimal selling schedule of each approximated orthogonal portfolio; third, weights of a neural network are pre-trained with the strategy suggested by previous step; last, we train the neural network to optimize on the original trading model.","summary":"Optimal execution of a portfolio have been a challenging problem for institutional investors. Traders face the trade-off between average trading price and uncertainty, and traditional methods suffer from the curse of dimensionality. Here, we propose a four-step numerical framework for the optimal…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Trades affect the market with a temporary component and a permanent component.","type":"execution","stated":"explicit","quote":"We assume trades affect the market with a temporary component and a permanent one.","support":"full"},{"assumption":"The exact functional form of temporary and permanent impact costs is not restricted.","type":"model","stated":"explicit","quote":"However, we do not rule the exact form of either.","support":"full"},{"assumption":"Asset returns in each period are normally distributed with parameters that depend only on the current regime.","type":"statistical","stated":"explicit","quote":"We assume the returns in each period follows a normal distribution whose parameters only depend on underlying regime.","support":"full"},{"assumption":"Regime switching occurs only at the end of each period and follows a Markov process.","type":"model","stated":"explicit","quote":"Regime switching happens only at the end of each period, and follows a Markov process.","support":"full"},{"assumption":"The current regime is directly observable to the trader, but future regimes cannot be foreseen.","type":"market","stated":"explicit","quote":"We assume the regime is directly observable to the trader, but she cannot foresee the future regime.","support":"full"},{"assumption":"The analysis focuses solely on the selling case of a large portfolio.","type":"execution","stated":"explicit","quote":"For simplicity, we only discuss the selling case in this paper.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Synthetic 10‑asset liquidation example with quadratic impact costs","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":"10 assets","quote":"The return of the risky assets under Regime 1 follows a multi-variate normal dis-tribution N([0.01, 0.005, 0.005]T , 10−5 ∗   5 1 1 1 3 1 1 1 3  ), and that under Regime 2 follows N([−0.005, 0.00, −0.01]T , 10−5 ∗   8 1 1 1 5 1 1 1 5  ). Given the parameters, we create the approximated orthogonal portfolios based on the permanent transaction cost matrices, as they are a better description of market impact","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"synthetic"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"Both temporary and permanent market impact costs are modeled; for illustration quadratic convex functions are used.","quote":"The market impact costs are modelled with a temporary part and a permanent part, where the former affects only the current trade while the latter persists. For illustration reasons, we present results with a quadratic convex market impact functions, but the impact can be any function of the trading amount.","source":"front_matter","rule":"R3","candidates":{"front_matter":"yes","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Both temporary and permanent market impact costs are modeled; for illustration quadratic convex functions are used.","card_quote":"The market impact costs are modelled with a temporary part and a permanent part, where the former affects only the current trade while the latter persists. For illustration reasons, we present results with a quadratic convex market impact functions, but the impact can be any function of the trading amount.","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"The combined method (dynamic program + neural network) outperforms the equal‑trade‑per‑period benchmark in both CRRA and mean‑variance objectives.","kind":"performance","quote":"The trading schedules suggested by the approximated orthogonal portfolio approach, though suboptimal, beats the benchmark of equal trades per period in many cases, and provides an advanced starting point for further machine learning improvement. The result outperforms the benchmark, and is tested to be robust.","support":"full"},{"finding":"Neural‑network refinement improves expected terminal wealth from $159.919 (orthogonal‑portfolio alone) to $160.829.","kind":"performance","quote":"We initilize the neural network with pre-trained weights, and train 1,000 steps with Adam Optimizer based on gradient method, resulting in a selling strategy with expected terminal wealth $160.829 and expected util-ity -0.00622. Comparing to the average wealth $159.919 and expected utility -0.00626 from approximated orthogonal portfolios alone, we notice a fair improvement by utilizing the neural networks.","support":"full"},{"finding":"Running time of the combined method scales linearly with the number of risky assets and trading periods.","kind":"performance","quote":"The running time is linear in the number of risky assets in the portfolio as well as in the number of trading periods. The running time of this combined method is expected to be linear in number of assets as well as in number of regimes, and can be even shortened by allowing parallel computing.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":12,"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":"2026-11-14T20:02:29Z"},"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/"}