Paper: arXiv 2509.22896
Authors: Peng Xu
Abstract
This paper develops stochastic optimization problems for describing and analyzing behavioral investors with Markowitz Stochastic Dominance (MSD) preferences. Specifically, we establish dominance conditions in a discrete state-space to capture all reverse S-shaped MSD preferences as well as all subjective decision weights generated by inverse S-shaped probability weighting functions. We demonstrate that these dominance conditions can be admitted as linear constraints into the stochastic optimization problems to formulate computationally tractable mixed-integer linear programming (MILP) models. We then employ the developed MILP models in financial portfolio analysis and examine classic behavioral factors such as reference point and subjective probability distortion in behavioral investors’ portfolio decisions.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 4.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper is highly mathematical, featuring advanced stochastic optimization, stochastic dominance theory, and detailed proofs leading to MILP formulations. However, its empirical section appears limited to applying the derived models to financial data for behavioral factor analysis, with no mention of actual backtesting, performance metrics, or implementation details, placing it in the research/lab phase.
Research Flowchart
flowchart TD A["Research Goal:<br>Model Behavioral Portfolio Decisions<br>using Markowitz Stochastic Dominance (MSD)"] --> B["Methodology:<br>Develop Stochastic Optimization with<br>MSD Constraints"] B --> C["Data/Input:<br>Asset Returns with Subjective<br>Probability Weighting"] C --> D["Computational Process:<br>Formulate as Mixed-Integer<br>Linear Programming (MILP)"] D --> E["Analysis:<br>Simulate Portfolio Choices<br>varying Reference Points & Probability Distortions"] E --> F["Key Findings/Outcomes:<br>1. Validated Reverse S-shaped Preferences<br>2. Quantified Impact of Reference Points<br>3. Demonstrated Tractable MILP for<br>Behavioral Portfolio Theory"]
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 novel stochastic portfolio optimization models that incorporate Markowitz Stochastic Dominance (MSD) and Modified MSD constraints as linear constraints, resulting in mixed‑integer linear programming (MILP) formulations.
Source sentence
We demonstrate that these dominance conditions can be admitted as linear constraints into the stochastic optimization problems to formulate computationally tractable mixed-integer linear programming (MILP) models.
Assumes 5
- The underlying state‑space S is discrete, consisting of n mutually exclusive and collectively exhaustive states. model
Source sentence
We assume that the underlying state-space S is discrete, consisting of n mutually exclusive and collectively exhaustive states, i.e., S = {s1, . . . , sn}
- States are indexed in ascending order of the benchmark portfolio Y’s returns. model
Source sentence
we also assume for the remainder of this research that the states {s1, . . . , sn} in S are indexed in ascending order of the state-specific returns of the benchmark portfolio Y
- The reference point r is chosen from one of the benchmark’s state‑specific returns. model
Source sentence
the reference point r is chosen from one of the benchmark’s state-specific returns, namely r ∈{yi | i ∈N}
- Portfolio returns X and Y are modeled as random variables supported on a common interval [a, b]. model
Source sentence
we consider a pair of portfolios whose returns are modeled by random variables X and Y supported on the interval [a, b] ⊂R
Show 1 more
- Incomplete preference information on utility functions and probability weighting functions is allowed. other
Source sentence
develop portfolio optimization models utilizing the MSD and MWSD criteria to accommodate incomplete preference information on both utility functions and probability weighting functions (PWFs)
Data
1964-01 – 2024-12 · US industry portfolios · equities · monthly
- 49 value‑weighted industry portfolios market-prices public
Source sentence
... includes all monthly return observations from Jan/1964 – Dec/2024, covering a sample period of 61 years, or 732 trading months.
- All‑share index (benchmark market portfolio) benchmark public
Source sentence
we utilize monthly return observations of the all‑share index from the Center for Research in Security Prices (CRSP) to proxy the benchmark market portfolio Y
Costs
Unclear - sources disagree
Finds 3
- The benchmark market portfolio is inefficient under both MSD and MWSD criteria, so it cannot be justified as optimal for behavioral investors with these preferences. theoretical
Source sentence
our empirical analysis reveals that the benchmark market portfolio is MSD- and MWSD-inefficient and thus cannot be justified as an optimal investment alternative for behavioral investors with MSD and MWSD preferences.
- MSD and MWSD investors achieve almost equal portfolio returns, but MWSD investors diversify more broadly across assets, especially when their returns are slightly lower. performance
Source sentence
MSD and MWSD investors appear to hold largely different portfolios in quite many 3‑year periods, although both achieve almost equally on portfolio returns. In general, MWSD investors have to diversify more broadly across the underlying asset universe ...
- Choice of reference point (risk‑free rate vs. benchmark median) and subjective probability distortion have little to no influence on MSD and MWSD portfolio returns. statistical
Source sentence
classic behavioral factors such as the choice of reference point (risk‑free rate or benchmark median) and subjective probability distortion report little to no influence on MSD and MWSD portfolio returns.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 32 related papers). Coverage is partial, so this is neither replication nor refutation.
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.