Papers for

financial engineers

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Cvar deep learning improves risk-aware decision timing in finance

CVaR Deep Optimal Stopping

Abstract: Risk-sensitive optimal stopping is useful when poor stopping outcomes carry substantial costs, as in option exercise, sequential medical treatment, and safety-critical control. A natural objective is Conditional Value at Risk (CVaR) of the stopped reward. We develop a theoretical framework and algorithm for CVaR deep optimal stopping. The framework and algorithm address four difficulties in applying risk-sensitive reinforcement learning to optimal stopping: state augmentation, score-function pathology, blindness to success, and dependence on auxiliary networks. We isolate three structural properties of optimal stopping: single-point reward, action-independent dynamics, and sum-zero gradient. We exploit these properties to establish Markov optimality without state augmentation and derive a pathwise gradient estimator. Under bounded policy derivatives, the pathwise estimator has bounded second moments near deterministic policies, avoiding the divergence that can occur with score-function estimators. Building on this framework, we propose DIOS (Distribution-Informed Optimal Stopping) to address the remaining two difficulties. A per-time-step soft penalty combined with tail-level annealing mitigates blindness to success. Estimating the reward quantile directly from each batch removes the need for an auxiliary critic. For a fixed tail level, we establish a finite-time stationarity guarantee under regularity assumptions along the training iterates. Across the 24 main Bermudan max-call configurations, DIOS achieves the highest mean CVaR and the shortest recorded training time among the tested CVaR baselines.

Sun 8 NovMachine Learning
The gist
Choosing the best time to stop an action can be risky if a bad choice has high costs, like in financial options or medical treatments. The authors study a way to focus on the worst-case outcomes using a measure called Conditional Value at Risk (CVaR). They designed a new method that learns how to stop optimally by fixing key problems in earlier techniques. Their method, called DIOS, learns faster and achieves better risk-aware decisions than previous methods on financial option tests.
Open → 2611.11332v1

Forecast augmented policy improves portfolio trading decisions

Foresee to Trade: Forecast-Augmented Policy Optimization for Portfolio Management

Abstract: Online investment services must translate asset opportunities into investment outcomes for users' accounts, yet the same market opportunity can warrant different account actions. We study how supervision from future asset opportunities can strengthen account-level trading decisions and propose Forecast-Augmented Relative Policy Optimization (FRPO). FRPO separately supervises asset opportunities and account decisions through a jointly learned representation: a Financial State Head (FSH) learns future relative-return targets to provide asset-level opportunity supervision, while Same-Path Group-Relative Optimization compares trading trajectories under the same market path and initial account to provide account-level relative-outcome supervision. An Adaptive Gate coordinates auxiliary supervision using predictive variance and gradient alignment. FSH is used only during training; inference uses the encoder and policy for trading. Across DJIA30, TOPIX Core30, and three investment-style portfolios, ten-seed evaluations show that FRPO achieves the highest mean cumulative and annualized returns among the compared methods in all five settings. These results support strengthening investment agents' trading decisions through joint supervision from asset opportunities and trading outcomes.

Thu 5 NovMachine Learning
The gist
Investment services need to pick the best trades for different user accounts even when market opportunities look the same. The authors introduce a method that teaches trading policies by combining predictions about future asset returns and comparing different trading decisions along the same market path. Their approach helps improve decision-making by using extra training signals related to both asset prospects and trading outcomes. The method was tested on multiple stock portfolios and consistently achieved higher returns than other techniques. It shows that training with future market information and outcome comparisons leads to better trading policies.
Open → 2611.07750v1

Equity portfolio method predicts and adjusts holdings to reduce worst losses

Interpretable Allocation and Model-Predictive Rebalancing of Equity Portfolios, Tested on Real and Synthetic Markets

Abstract: A live equity portfolio poses three problems: which stocks to hold, when to trade, and whether the strategy holds up beyond the observed market. We built one system for all three and tested it on 373 US equities, calibrated on 2014-2024 and traded from January 2025 to September 2026. A selection rule chose stocks by score, by a bandit, or at random, and one of seven weighting rules, four consumer-theory utilities, and three standard portfolios, set share counts under limits on budget, single-stock weight, portfolio volatility, and turnover, with each stock's preference weight a bounded score of its trend, market beta, and the market's recent direction. A model-predictive controller forecasts portfolio value with a multi-asset geometric Brownian motion and re-planned at its horizon, when the value left its forecast band, or at a drawdown stop. To probe deeper declines than the real window held, we also ran the system on simulated markets from per-asset hidden Markov generators with jumps, linked by market and sector factors. On real prices, the stocks chosen mattered more to terminal wealth than the weighting rule, and the early re-plans shrank the worst drawdown in most rule combinations against a fixed calendar without consistently raising wealth. In the deepest simulated declines, every combination, random picks included, ended above simulated SPY and the equal-weight universe in the median, even without early re-plans, so the margin came from shared parts of the system, not from stock selection or re-plan timing. A market hedge that tightened a beta cap as the market weakened cost wealth on real prices but paid off in the most sustained simulated declines, and the generator's persistent trends favored score-based selection more than the real market did, which suggests a synthetic test judges stock selection fairly only if its generator does not build in that persistence.

Mon 2 NovComputational Engineering, Finance, and Science
The gist
Managing an equity portfolio involves choosing which stocks to hold, deciding when to trade, and checking if the strategy works in different markets. The authors created a system that picks stocks using different rules and adjusts holdings by forecasting portfolio values through a mathematical model. They tested this system on real and simulated markets, finding that stock choice had more impact on final wealth than how shares were weighted, and early adjustments helped reduce large losses without always increasing wealth. In extreme simulations, the system outperformed standard indexes even with random stock picks, showing its built-in approach contributed most to success.
Open → 2611.03022v1

High dimensional integration improved by domain confined learned samplers

Numerical Integration on Bounded Domains via Learned Reflected Diffusions

Abstract: We study high-dimensional integration on bounded convex domains. Given a non-negative integrand $r$, the goal is to estimate $Z=\int_{D}r(x)\,\mathrm dx$ on $D\subset\mathbb R^d$. Standard Quasi-Monte Carlo does not adapt its evaluation points to the integrand, while effective importance sampling requires a proposal that captures its concentration. Learned diffusion samplers can supply such a proposal, but existing methods are formulated on $\mathbb R^d$ and do not respect the domain constraint $x\in D$. We instead build the proposal from reflected dynamics that stay in $D$ by construction. A reflected diffusion progressively transforms the target distribution toward the uniform distribution on $D$, and a neural score model is trained to reverse this process. A change-of-measure formula for reflected diffusions yields a training objective without additional boundary terms and an unbiased importance-sampling estimator of $Z$. For non-smooth integrands, where reparameterization gradients may be undefined or uninformative, we derive a REINFORCE estimator of the training gradient using Gaussian proposal densities before reflection, without evaluating reflected transition densities or differentiating $r$. We evaluate on numerous synthetic targets and real-world problems, demonstrating strong performance of the proposed method, as well as favorable scaling with the dimension of the problem and the number of points used for the Monte Carlo estimate.

Thu 29 OctMachine Learning
The gist
High-dimensional integration means finding the total value of a complicated function across a certain space, which is very hard when the space is limited or shaped unusually. The authors propose a new way to sample points inside such bounded spaces using a process called reflected diffusion, which keeps points inside the allowed region by 'bouncing' off the boundaries. They teach a neural network to reverse this diffusion so it can produce better samples that focus on the important areas of the function. This method helps calculate integrals more accurately and efficiently, especially in many dimensions.
Open → 2610.40499v1

Deep learning method improves control of systems with random jumps

Deep Pontryagin Learning for Continuous-Time Stochastic Control with Jumps

Abstract: We develop a deep Pontryagin learning framework for high-dimensional continuous-time stochastic control problems that naturally accommodates jump-diffusions. In contrast to direct policy optimization, which repeatedly differentiates the objective through the entire time-discretized dynamics, our method separates adjoint learning from policy improvement by explicitly incorporating state-sensitivity information from the Pontryagin stochastic maximum principle. For a fixed Markov control, we first compute pathwise adjoint labels by backpropagating the reward through time along simulated state trajectories, and project these labels onto the Markov state through a supervised regression step. The control is then updated by enforcing the pointwise Hamiltonian stationarity condition. We prove that the resulting projected adjoint process satisfies a discrete-time Pontryagin principle and converges, as the time step tends to zero, to the solution of the corresponding continuous-time adjoint backward stochastic differential equation. Numerical experiments illustrate the accuracy, stability, and scalability of the approach.

Mon 19 OctMachine Learning
The gist
Controlling complex systems that can suddenly change in unpredictable ways is hard, especially when these changes happen over continuous time. The authors introduce a deep learning framework that separates understanding how small changes affect the system from improving control decisions. Their method uses mathematical tools called the Pontryagin principle to guide learning more efficiently. Tests with simulations show this approach is accurate, stable, and works well for large problems.
Open → 2610.26718v1

Momentum methods improve nested stochastic optimization efficiency

Optimal Momentum Methods for Stochastic Multilevel Compositional Optimization

Abstract: This paper investigates stochastic multi-level optimization where the objective is a nested composition of several smooth non-convex functions. We assume that only stochastic estimates of the gradient and function values for each level are accessible. Consequently, obtaining an accurate estimate of the overall gradient is challenging due to the nested structure. To address this, we employ a momentum-based estimator with mini-batches to track the function values of each level, which are subsequently used to construct momentum gradient estimators. We establish an optimal sample complexity of $\mathcal{O}(ε^{-4})$ for finding an $ε$-stationary point, avoiding the stronger average smoothness assumption commonly relied upon in prior literature. Furthermore, by employing a normalization technique, we attain the same rate without requiring problem-dependent constants to set hyperparameters. To achieve the optimal rate without mini-batches, we further develop a batch-free method that incorporates a first-order approximation and a clipping technique for function value estimation. Finally, we validate the effectiveness of our proposed methods through experiments on risk-averse portfolio optimization and hierarchical tilted empirical risk minimization.

Thu 1 OctMachine Learning
The gist
Optimizing problems made up of several layers of functions can be tricky because you only get noisy estimates of the steps needed to improve them. The authors develop smarter ways to combine information from each layer using momentum, which helps speed up finding good solutions. Their methods work well without needing some strong assumptions that others required before. They tested these techniques on financial portfolio choices and risk-sensitive prediction tasks, showing their approach works in practice.
Open → 2610.01572v1

Functional dynamic mode decomposition learns infinite dimensional systems from data

Functional dynamic mode decomposition: Learning infinite-dimensional systems from data

Abstract: Dynamic mode decomposition (DMD) is a data-driven method that computes the best linear approximation of the underlying dynamical system and decomposes the dynamics into a superposition of characteristic spatiotemporal patterns. Originally introduced by the fluid dynamics community, DMD and its extensions have found widespread use in many other research areas such as molecular dynamics, climate science, engineering, finance, and neuroscience. Applications include dimensionality reduction, forecasting, system identification, control, and spectral clustering. In order to apply DMD to partial differential equations, the spatial domain is typically first discretized using finite difference or finite element techniques, thus implicitly rendering the problem finite-dimensional. We extend projected and exact DMD to infinite-dimensional systems. Rather than estimating matrices from vector-valued observations, our DMD variants learn finite-rank operators from functional data such as observables, densities, or wavefunctions. We show that conventional DMD algorithms can be regarded as special cases of their functional DMD counterparts. All results will be illustrated with the aid of guiding examples. We focus in particular on Koopman, Perron-Frobenius, and Koopman-von Neumann operators associated with graphons, ordinary differential equations, and stochastic differential equations.

Thu 24 SeptMachine Learning
The gist
Dynamic mode decomposition (DMD) is a technique to simplify and understand complex changing systems by breaking them into patterns. The authors extend DMD to work directly with infinite-dimensional data, like continuous functions, rather than limiting it to simpler finite data. This lets them analyze systems modeled by equations involving continuous spaces without approximating them as finite sets. Their approach connects to important mathematical operators used to describe dynamics in areas like networks and random processes, helping to generalize DMD for broader scientific use.
Open → 2609.29159v1