When Minimizing Risk Allows Zero Wealth

TL;DR for operators

A portfolio-risk team can minimize a downside-risk measure while leaving the most severe losses outside its protection. The critical design choice is which probability-ranked outcomes enter the objective.

In Dynamic Portfolio Selection under Range Value-at-Risk with Logarithmic Returns, Hui Mi, Yuzhe Pan, and Zuo Quan Xu (2026)¹ show that pure minimization of Range Value-at-Risk (RVaR) can assign zero terminal wealth to sufficiently adverse market states when the measure excludes the lowest return quantiles. Their main theoretical contribution identifies two distinct optimal-payoff structures and derives corresponding dynamic strategies.

The numerical implications are equally relevant to portfolio oversight. In one stylized Black–Scholes calibration, pure RVaR minimization produces an initial risky-asset position of 3.5974 times initial wealth, exceeding the growth-optimal benchmark of 2.6667. Exposure also responds non-monotonically to optimization weights.

For operators, the implication is to examine excluded-tail outcomes, terminal-wealth floors, and leverage sensitivity independently of the optimized risk score. The paper establishes these properties mathematically; it does not demonstrate improved performance in actual markets.

A downside measure can exclude the most adverse losses

Portfolio managers commonly choose between measuring losses at a particular probability threshold and averaging losses across a broader part of the return distribution. Averaging sounds more informative because it accounts for multiple outcomes. Yet an average only represents the observations included in its calculation.

RVaR allows the manager to choose a lower quantile endpoint, $\alpha$, and an upper endpoint, $\beta$, and average the corresponding return quantiles:

Here, $G_R(u)$ is the return quantile function, and $0\leq\alpha<\beta\leq1$.

When $\alpha=0$, the interval includes the lowest returns and RVaR becomes Expected Shortfall (ES) at $\beta$. As $\beta$ approaches $\alpha$ from above, RVaR approaches Value-at-Risk (VaR).

When $\alpha>0$, however, the most adverse $\alpha$ fraction of returns is excluded. For example, $\alpha=0.2$ and $\beta=0.21$ evaluate the interval between the 20th and 21st return percentiles, not the worst 20%.

This creates an unusual incentive when RVaR becomes an optimization objective. Losses below the lower endpoint do not directly worsen the measured criterion. Under pure RVaR minimization, the optimizer can therefore sacrifice those outcomes to improve wealth elsewhere.

In the paper’s optimal solutions, this possibility is explicit: zero terminal wealth appears in sufficiently adverse states. Because performance is measured using logarithmic return, zero wealth also implies negative-infinite logarithmic return.

The result is not evidence of a numerical solver malfunction. It follows from the objective being optimized.

Two payoff geometries explain the zero-wealth result

The paper’s main contribution is an analytical characterization of the optimal portfolio, not simply the observation that truncated risk measures can overlook extreme losses.

The authors consider an investor balancing expected logarithmic growth against RVaR:

where $R_T=T^{-1}\log(X_T/x)$, $\lambda$ controls the growth preference, and terminal wealth must satisfy the market’s budget constraint.

In a complete diffusion market, different future states have different present costs for delivering wealth. The state-price density captures those costs. Higher wealth can be allocated to less expensive states while preserving the same overall terminal-wealth distribution.

The authors use this relationship to convert the continuous-time investment problem into a static quantile optimization problem.

There is a further mathematical complication: a valid wealth quantile function must be nondecreasing, but RVaR’s weighting across a selected interior interval can generate a transformed objective whose slopes do not satisfy the required ordering.

The solution uses a convex envelope: the greatest convex function lying below the transformed weighting function. Its derivative produces an appropriately ordered optimal wealth quantile function. Linear envelope segments create constant terminal-payoff regions, while other segments produce wealth proportional to the inverse state-price density.

The resulting geometry has two scenarios:

These are the paper’s central analytical results (Section 3.3, Propositions 3.7–3.9). The scenario distinction depends on supporting-line contact points in the convex-envelope construction, rather than a simple, universally applicable economic threshold.

The authors also derive replicating wealth processes and risky-asset positions under constant investment opportunities (Corollaries 3.8 and 3.10). A constant terminal payoff region should not be interpreted as a continuously risk-free position: implementing a state-contingent terminal payoff can require dynamic trading.

Finally, the VaR and ES limiting solutions recover the corresponding portfolio structures (Section 3.4, Propositions 3.11–3.12). These are analytical consistency checks, not empirical comparisons demonstrating one risk measure’s superiority.

More emphasis on risk does not guarantee less leverage

The paper’s numerical sensitivity analysis illustrates how these payoff structures translate into initial risky-asset positions.

Its Black–Scholes setting uses unit initial wealth, a one-period horizon, a 5% risk-free rate, 15% volatility, and a market price of risk of 0.4. The growth-optimal benchmark invests 2.6667 monetary units in the risky asset for every unit of initial wealth.

For $\alpha=0.2$ and $\beta=0.21$, Table 1 reports:

The position initially falls as growth receives positive weight, then rises, falls again, and eventually returns to the growth-optimal benchmark.

A broader RVaR interval is not necessarily associated with lower leverage either. At $\lambda=0.2$ and $\alpha=0.2$, increasing $\beta$ from 0.21 to 0.7 raises initial risky investment from 2.6455 to 3.1902.

These are model-implied sensitivity calculations, not a historical backtest. Their purpose is to demonstrate non-monotone comparative statics: $\alpha$, $\beta$, and $\lambda$ jointly affect the geometry of the optimal payoff, so no single parameter reliably determines portfolio aggressiveness.

Portfolio governance must cover outcomes outside the objective

For asset managers, Cognaptus draws three operational implications from the analysis.

First, portfolio-risk teams selecting RVaR endpoints should document which return quantiles enter the objective and which remain excluded. When $\alpha>0$, diagnostics covering the lower excluded tail are necessary to assess potential catastrophic outcomes.

Second, model validators should inspect optimal terminal-wealth distributions alongside RVaR values. A low reported risk score is insufficient if the solution permits zero terminal wealth. Minimum-wealth requirements or independent downside constraints are possible safeguards, although this paper does not test their effectiveness.

Third, portfolio managers and investment committees should recalculate risky exposure across a grid of $\alpha$, $\beta$, and $\lambda$. Treating a higher risk weight or a wider evaluation interval as automatically more conservative can conceal significant changes in leverage.

For quantitative investment engineers, the paper’s analytical payoff and replication formulas also provide reference cases against which numerical implementations can be checked. That validation role is particularly relevant when an optimizer produces unexpectedly large positions.

Analytical optimality is not deployment readiness

The mathematical characterization depends on a complete diffusion market, regular state-price-density distributions, and continuous self-financing trading. Transaction costs and explicit portfolio constraints are absent. Closed-form dynamic replication additionally relies on constant investment opportunities.

The numerical evidence comes from a stylized Black–Scholes calibration. There is no observed return dataset, out-of-sample evaluation, execution analysis, or evidence that the resulting strategies outperform VaR- or ES-based portfolios in realized risk or return.

Accordingly, the paper establishes how RVaR optimization behaves under specified assumptions, while the consequences of trading frictions, constrained leverage, incomplete markets, and estimation errors remain unresolved.

The risk objective is not a capital-survival guarantee

Mi, Pan, and Xu show that selecting a downside-risk interval changes more than a reported risk statistic. It changes the mathematical structure of the optimal portfolio and the wealth assigned to different market outcomes.

For investment teams, the central lesson is that formal optimization can be correct even when the resulting loss profile violates the institution’s risk priorities. The risk measurement window, allowable terminal losses, and acceptable leverage must be governed together if the objective is to protect capital beyond the quantiles explicitly measured.

Cognaptus: Automate the Present, Incubate the Future.