📈 economics

To Gamble, Perchance to Grow

This paper establishes that a return transformation yields a more conservative growth-optimal portfolio if and only if the function is concave, strictly increasing, and satisfies a specific convexity condition, thereby characterizing comparative risk aversion for rationally inattentive agents.

Original authors: Mark Whitmeyer

Published 2026-06-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mark Whitmeyer

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The "Growth-Optimal" Investor

Imagine an investor who plays the long game. Their only goal isn't to get rich quick or to avoid losing a single dollar; their goal is to make their wealth grow as fast as possible over a lifetime. In the world of economics, this is called the Kelly Investor.

This investor faces a simple choice every day:

The Safe Bet: A guaranteed return (like a savings account).
The Risky Bet: A gamble that could pay out a lot or a little (like a stock).

The paper asks a specific question: What happens if we change the rules of the game by "transforming" the returns?

Imagine the returns are numbers on a scoreboard. The author asks: If we apply a mathematical filter to every number on that scoreboard (making big numbers smaller, or small numbers bigger), will the investor suddenly become more cautious and put less money into the risky gamble?

The Core Discovery: It's Not Just About "Smoothing"

Most people think that if you make a risky payoff curve "smoother" or "flatter" (mathematically, making it concave), the investor will naturally become more risk-averse. Think of it like putting a cushion under a trapeze artist; they feel safer, so they might take fewer risks.

The paper says: Not so fast.

The author proves that simply "cushioning" the returns (making the function concave) is not enough to make the Kelly investor universally more conservative. You can have a very smooth, safe-looking curve, and the investor might still decide to gamble more.

To guarantee the investor becomes more cautious, the transformation must satisfy two conditions simultaneously:

It must be "concave": It compresses the upside. (The cushion is there).
It must satisfy a "Harmonic Curvature" rule: This is the tricky part. The author introduces a second condition involving the ratio of the return to the transformed return ( $r/f(r)$ ).

The Analogy of the "Double-Check":
Imagine you are a judge deciding whether to let a driver speed up.

Condition 1 (Concavity): You check if the road has potholes that might slow them down.
Condition 2 (Harmonic Curvature): You check if the relationship between the speed limit and the actual speed has changed in a specific way.

The paper argues that you need both checks to pass. If you only have the first check (concavity), the driver might still speed up. The second check ensures that the "cost" of the risk feels heavy enough in the transformed world to make the driver hit the brakes.

The "Mirror" Effect

The paper also looks at the opposite scenario: What makes an investor more willing to gamble?

To make an investor more conservative, the transformation must be concave AND the ratio $r/f(r)$ must be convex.
To make an investor more aggressive, the transformation must be convex AND the ratio $r/f(r)$ must be concave.

It's like a mirror image. The math requires a specific balance between how the numbers are squashed and how they stretch relative to each other.

Real-World Examples from the Paper

The author gives concrete examples to prove that "just being concave" isn't enough:

The "Safe" Transformation: If you transform returns using $f(r) = r / (1+r)$ , the investor becomes more conservative. This function is concave, and it passes the second "harmonic" test.
The "Trap" Transformation: If you use $f(r) = \sqrt{r}$ (the square root), it is concave (it flattens the curve). However, it fails the second test. In this specific case, the investor actually becomes less conservative and puts more money into the risky asset, even though the curve looks "safer."

The "Rational Inattention" Connection

The paper also connects this to a concept called Rational Inattention. Imagine an agent who is smart but has a limited attention span (like a human with a busy brain). They have to decide between a safe action and a risky action, but processing the information about the risky action costs "mental energy."

The author shows that the math for this "tired brain" is the same as the math for the "growth investor."

If you transform the utility (the happiness) of the agent, they only become more risk-averse if the transformation is "sufficiently concave."
It's not enough to just make the utility curve bend down; it has to bend down hard enough to overcome the cost of paying attention.

The "Outside Option" Surprise

Finally, the paper looks at a scenario where, after the gamble is over, the investor gets a "safety net" or an outside option.

The Setup: You invest in a risky asset. If it pays out $10, great. If it pays out $0, you get a guaranteed $5 from an outside source.
The Result: Surprisingly, having this safety net makes the investor more willing to take the risk.
The Analogy: Think of a tightrope walker. If you tell them, "If you fall, you won't hit the ground; you'll land on a trampoline," they might actually walk the tightrope faster or take more dangerous stunts because the downside is capped. The paper proves that adding an "outside option" (a floor under the worst outcomes) universally encourages more gambling.

Summary

Goal: To find which mathematical changes to investment returns make a growth-focused investor always play it safer.
Finding: Simply making returns "safer" (concave) isn't enough. You need a specific second condition (harmonic curvature) to ensure the investor actually reduces their risk.
Insight: A "safety net" (outside option) paradoxically makes people more willing to gamble, not less.
Takeaway: In the world of optimal growth, intuition about "safety" can be misleading. You need a precise mathematical balance to guarantee a change in behavior.

Technical Summary: To Gamble, Perchance to Grow

Problem Statement
This paper investigates the conditions under which a deterministic or random transformation of asset returns induces a more conservative investment behavior in the growth-optimal (Kelly) portfolio problem. Specifically, the author considers an investor maximizing expected log-growth with access to one safe asset (gross return $s$ ) and one risky asset (gross return $R$ ). The central question is: which transformations $f$ of the return space $(0, \infty) \to (0, \infty)$ universally increase the demand for the safe asset (and conversely, decrease the demand for the risky asset) for every risk-free return $s$ and every finite-support distribution of the risky return?

The paper frames this inquiry through the lens of comparative risk aversion. While classical literature (Pratt, 1964; Ross, 1981; Kimball, 1990) defines risk aversion via transformations of utility functions, this study asks which transformations of returns make a Kelly investor appear more risk-averse. The author interprets "more risk-averse" in the behavioral sense of Yaari (1969): an agent is more risk-averse if they choose a lower share of the risky asset.

Methodology
The analysis proceeds by normalizing the problem. Let $Y = R/s$ be the risky return relative to the safe return. The investor solves:
$\max_{\alpha \in [0,1]} E_p \log(1 + \alpha(Y - 1))$
where $\alpha$ is the share of wealth in the risky asset. A transformation $f$ induces a new relative return transformation $T_s^f(y) = f(sy)/f(s)$ . The transformed problem becomes maximizing expected log-growth under $T_s^f(Y)$ .

The core analytical tool is the score amplification ratio, defined for a portfolio share $\alpha$ and relative return $y$ as:
$k_{\alpha,s}^f(y) = \frac{\gamma_\alpha(T_s^f(y))}{\gamma_\alpha(y)}$
where $\gamma_\alpha(y) = \frac{y-1}{1+\alpha(y-1)}$ is the marginal log-growth gain.

The author establishes that universal safe-demand monotonicity is equivalent to a single-crossing property of this score ratio: the transformation must amplify the negative marginal incentives of "bad" realizations ( $y < 1$ ) at least as much as it amplifies the positive marginal incentives of "good" realizations ( $y > 1$ ). By analyzing the endpoints of the portfolio share ( $\alpha=0$ and $\alpha=1$ ), the author reduces this dynamic condition to static curvature properties of the transformation function $f$ .

The methodology extends to:

Rational Inattention: Leveraging the equivalence between the Kelly problem and the rational-inattention problem (Robson et al., 2023) to characterize comparative risk aversion for agents with Shannon information costs.
Random Transformations: Analyzing families of transformations where the investor faces an ex-post outside option (e.g., $r \vee B$ ), treating the transformation as a random variable independent of returns.

Key Contributions and Results

1. Deterministic Characterization (Theorem 3.3)
The paper provides a necessary and sufficient condition for a transformation $f$ to universally raise safe demand. A transformation $f$ raises safe demand universally if and only if:

$f$ is strictly increasing.
$f$ is concave.
The function $r \mapsto r/f(r)$ is convex.

The author emphasizes that ordinary concavity is insufficient. The third condition, termed a "second harmonic curvature restriction," requires that the transformation compresses upside returns sufficiently relative to the downside.

Example: $f(r) = r/(1+r)$ satisfies these conditions (safe demand rises).
Counter-example: $f(r) = \sqrt{r}$ is concave but fails the convexity of $r/f(r)$ ; consequently, it does not universally raise safe demand (in some cases, it increases risky demand).

2. Rational Inattention Corollary (Corollary 4.3)
Applying the main theorem to the rational-inattention framework, the paper characterizes transformations $\phi$ of utility (growth rates) that make an agent more risk-averse. For a smooth transformation $\phi$ , the agent is more risk-averse (chooses the safe action more frequently) if and only if:
$\phi''(u) \leq -|\phi'(u)(1 - \phi'(u))|$
This result implies that ordinary concavity of $\phi$ is not enough; the new utility must be "sufficiently more concave" to overcome the scaling effects inherent in the information cost structure.

3. Random Transformations and Outside Options (Theorem 5.3 & Corollaries 5.4, 5.5)
The paper extends the analysis to random transformations $F_z$ . A key finding is that ex-post outside options (where the investor can take $\max\{r, B\}$ instead of $r$ ) universally raise risky demand.

The transformation $r \mapsto r \vee B$ satisfies the single-crossing condition for risky demand.
This holds for both the standard Kelly investor and the rationally inattentive agent, implying that the availability of an outside option makes the agent less risk-averse (more willing to hold the risky asset).

Significance and Claims
The paper claims to distinguish return transformations from utility transformations. While classical comparative risk aversion relies on concave transformations of utility, this study demonstrates that in the growth-optimal context, return transformations require a specific dual curvature condition (concavity of $f$ and convexity of $r/f(r)$ ) to induce conservative behavior.

The author positions the work as distinct from the portfolio-dominance literature (which alters payoff distributions) and standard portfolio-growth questions (which seek optimal strategies). Instead, the paper isolates the effect of cardinal return scales on portfolio shares. The findings suggest that "more risk-averse" behavior in a Kelly framework is not merely a matter of utility curvature but depends on how the return scale itself is warped, specifically requiring that the transformation makes original returns sufficiently convex when measured in transformed units.

The paper concludes modestly, noting that these results provide a precise characterization of when a Kelly investor behaves more conservatively, offering a new perspective on comparative risk aversion that is robust to all finite-support distributions of risky returns.