Paper: arXiv 2305.11319

Authors: Silvana M. Pesenti, Sebastian Jaimungal, Yuri F. Saporito, Rodrigo S. Targino

Abstract

We define and develop an approach for risk budgeting allocation - a risk diversification portfolio strategy - where risk is measured using a dynamic time-consistent risk measure. For this, we introduce a notion of dynamic risk contributions that generalise the classical Euler contributions and which allow us to obtain dynamic risk contributions in a recursive manner. We prove that, for the class of coherent dynamic distortion risk measures, the risk allocation problem may be recast as a sequence of strictly convex optimisation problems. Moreover, we show that self-financing dynamic risk budgeting strategies with initial wealth of 1 are scaled versions of the solution of the sequence of convex optimisation problems. Furthermore, we develop an actor-critic approach, leveraging the elicitability of dynamic risk measures, to solve for risk budgeting strategies using deep learning.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 3.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper develops a theoretical framework for dynamic risk budgeting using advanced concepts like coherent dynamic distortion risk measures, Gâteaux derivatives, and strictly convex optimization, leading to a high math complexity. However, the empirical implementation is limited to a conceptual deep learning actor-critic approach with illustrative examples, lacking detailed backtest results, code, or statistical metrics, resulting in low empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Risk Budgeting for Dynamic Risk Measures"] --> B["Methodology:<br>Define Dynamic Risk Contributions & Convex Optimization"]
  B --> C["Input:<br>Multi-Asset Portfolio Data"]
  C --> D["Process:<br>Actor-Critic Deep Learning Algorithm"]
  D --> E["Outcome:<br>Time-Consistent Dynamic Risk Budgeting Strategy"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a risk budgeting allocation framework using dynamic time‑consistent risk measures, introduces dynamic risk contributions, shows the problem can be solved via a sequence of strictly convex optimisation problems, and develops an actor‑critic deep‑learning algorithm leveraging elicit

Source sentence
We define and develop an approach for risk budgeting allocation - a risk diversification portfolio strategy - where risk is measured using a dynamic time-consistent risk measure. For this, we introduce a notion of dynamic risk contributions that generalise the classical Euler contributions and which allow us to obtain dynamic risk contributions in a recursive manner.

Assumes 4

  • Assets follow a discrete‑time Heston‑inspired market model with student‑t copula dependence. market
    Source sentence
    Market Model and NN Hyperparameters We use a discrete time version of a Heston inspired market model over a time horizon of T + 1 = 12 months, where asset returns have a student-t copula dependence. The investor updates their strategy monthly, corresponding to time index t ∈T in our optimisation problem.
  • The one‑step risk measures are coherent distortion risk measures satisfying positive homogeneity and full‑allocation (Euler‑like) properties. model
    Source sentence
    Let {ρt}t∈T be a coherent distortion DRM. Then it holds for any admissible strategy θ0:T that Rt[θt:T] = X i∈N RCt,i[θt:T], a.s., ∀t ∈T .
  • Risk contributions are the Gâteaux derivatives of a positive homogeneous risk‑to‑go function. statistical
    Source sentence
    For all t ∈T and for all at ∈L∞ t , at > 0, we have that at RCt,i [θt:T] = RCt,i [(at θt, θt+1:T)] = RCt,i [(at θt:T)] . (6) Proof: This follows as RCt,i are the Gˆateaux derivatives of a positive homogeneous function.
  • The risk‑to‑go and conditional CDF are elicitable functionals, allowing estimation via convex optimisation rather than nested simulation. statistical
    Source sentence
    5.2. Conditional Elicitability As the risk-to-go Rt[θt:T] = ρt (θ⊺ t∆Xt + wθ t Rt+1[θt+1:T]) consists of one-step (conditional) risk measures, both the risk-to-go and the conditional cdf could be estimated via nested simulations, which, however is numerically expensive. The key concept is that a functional is elicitable, e.g., mean, VaRα, ESα, if it is the minimiser of the expected value of a scoring func-tion.

Data

  • Synthetic Heston‑inspired market simulation synthetic unknown
    Source sentence
    Market Model and NN Hyperparameters We use a discrete time version of a Heston inspired market model over a time horizon of T + 1 = 12 months, where asset returns have a student-t copula dependence. The investor updates their strategy monthly, corresponding to time index t ∈T in our optimisation problem.

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • For coherent distortion DRM, risk contributions satisfy the full‑allocation property: total risk equals the sum of individual contributions at every time step. theoretical
    Source sentence
    Let {ρt}t∈T be a coherent distortion DRM. Then it holds for any admissible strategy θ0:T that Rt[θt:T] = X i∈N RCt,i[θt:T], a.s., ∀t ∈T .
  • The risk budgeting problem can be recast as a sequence of strictly convex optimisation problems, and self‑financing strategies with initial wealth 1 are scaled versions of these solutions. theoretical
    Source sentence
    We prove that, for the class of coherent dynamic distortion risk measures, the risk allocation problem may be recast as a sequence of strictly convex optimisation problems. Moreover, we show that self-financing dynamic risk budgeting strategies with initial wealth of 1 are scaled versions of the solution of the sequence of convex optimisation problems.
  • An actor‑critic deep‑learning algorithm using neural networks for the actor (strategy) and critics (conditional CDF and risk‑to‑go) successfully approximates the optimal risk budgeting strategies. theoretical
    Source sentence
    Approximation of Risk Budgeting Strategies While Theorems 3 and 4 provide a characterisation of risk budgeting strategies as solutions to a sequence of convex optimisation problems, they do not provide a methodology for finding them. The general strategy behind our actor-critic algorithm is: 1.)
  • Numerical experiments on the synthetic stochastic‑volatility market model demonstrate stability of the proposed deep‑learning scheme across different risk budgets. performance
    Source sentence
    Finally, we demonstrate the stability of the numerical scheme through several examples using a stochastic volatility market model. 28 Pesenti, Jaimungal, Saporito, Targino: Risk Budgeting Allocation for Dynamic Risk Measures equal bt,i p = 50% p = 90% 0 3 6 9 0.05 0.10 0.15 0.20 0.25 0.30 0.35 i = 1 i = 2 i = 3 i = 4 i = 5 0 3 6 9 0.05 0.10 0.15 0.20 0.25 0.30 0.35 i = 1 i = 2 i = 3 i = 4 i = 5 unequal bt,i p = 50%

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 15 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 9 papers in the archive (5 by independent authors)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.