The Long-Only Minimum Variance Portfolio in a One-Factor Market: Theory and Asymptotics

This paper provides an explicit solution and computable characterization for the long-only minimum variance portfolio under a one-factor model with mixed-sign betas, resolving a prior open question and establishing that in high-dimensional regimes, the proportion of active assets converges to a value determined by the distribution of asset betas, notably vanishing when all betas are positive.

The Gist

Imagine you are the captain of a massive ship (a portfolio) sailing through a stormy ocean (the stock market). Your goal is simple: keep the ship as steady as possible (minimize risk/variance) while ensuring you have enough fuel to reach your destination (invest 100% of your money).

In the world of finance, there are two ways to sail:

1. The "Long-Short" Ship: You can sail forward (buy stocks) or sail backward (short-sell stocks). This is like having a rudder that can turn both ways. It's very flexible, but in a storm, it can sometimes spin out of control or require complex maneuvers that real-world investors aren't allowed to do.
2. The "Long-Only" Ship: You can only sail forward. You can't short-sell. This is the rule for most real-world funds (pension funds, mutual funds) because short-selling is expensive, risky, and often forbidden.

This paper is about finding the perfect, steadiest path for that "Long-Only" ship, specifically when the ocean is driven by one giant wave (a "one-factor market," like the overall economy moving up or down).

Here is the breakdown of their discovery, using simple metaphors:

1. The Problem: Who Gets on the Boat?

When you try to build the steadiest "Long-Only" portfolio, you face a tricky puzzle. You have thousands of potential passengers (assets).

• In the "Long-Short" world, you might put 50% of your money in Apple, -20% in Tesla (betting against it), and 70% in Google.
• In the "Long-Only" world, you can't bet against Tesla. So, you have to decide: Which passengers get a seat on the boat, and which ones get left on the dock?

The authors found that you don't need to guess. There is a secret "cutoff line" based on a number called Beta (which measures how much a stock moves with the big wave).

• The Discovery: You simply line up all your assets from the "lowest wave-chaser" to the "highest wave-chaser."
• The Rule: You take everyone from the bottom of the list up to a specific point, and you leave everyone else behind.
• The Twist: Even if some stocks have "negative" betas (they move opposite to the wave), the math still works! The authors solved a puzzle that previous experts thought was impossible to solve rigorously when negative betas were involved.

2. The "One-Wave" Ocean (The One-Factor Model)

The authors assume the market is mostly driven by one giant wave (the general economy).

• The Analogy: Imagine a dance floor where everyone is dancing to the same beat. Some dancers are very energetic (high Beta), some are lazy (low Beta), and some are dancing in reverse (negative Beta).
• The Result: To keep the dance floor from shaking too much, you only want the dancers who are moving just enough to stay in sync, but not so much that they cause chaos. The paper gives a precise formula to find exactly how many dancers to keep.

3. The Big Surprise: The "Empty Boat" Phenomenon

Here is the most fascinating part of the paper. They looked at what happens when you have a massive number of assets (thousands or millions).

• The Old Belief: People thought that if you have thousands of stocks, you'd probably use most of them to diversify your risk.
• The New Reality: If the "wave" (the market) is strong and most stocks move in the same direction (positive betas), the "Long-Only" portfolio becomes extremely picky.
• The Metaphor: Imagine you have a basket of 1,000 apples. You want to pick the perfect ones to make a pie. If you are allowed to throw away bad apples (shorting), you might keep 500. But if you can only pick good apples (long-only), you might realize that only the top 5 apples are perfect enough to keep the pie stable. The rest are too risky to include.

The Math Magic:
The paper proves that as the number of assets grows, the percentage of assets you actually use (the "Active Ratio") often shrinks toward zero.

• If all stocks move with the market, you only need a tiny handful of the "calmest" stocks to build a safe portfolio.
• If a few stocks move against the market (negative betas), you might include a few more, but the number is still surprisingly small.

4. The "Cube Root" Rule

The authors also looked at what happens when the number of "bad apples" (negative betas) is very small.

• They found a "Cube Root" relationship. If the chance of finding a negative-beta stock is tiny, the number of stocks you need to include in your portfolio grows, but very slowly.
• Analogy: If you have a needle in a haystack, finding one needle doesn't mean you have to dig through the whole haystack. You only need to dig a little bit deeper. The math shows that even if you double the number of "risky" stocks, your portfolio size doesn't double; it grows much more gently.

Summary: What does this mean for you?

1. Simplicity Wins: You don't need a supercomputer to find the best "Long-Only" portfolio. There is a clear, step-by-step rule (sort by Beta, cut off at a specific line) that works even if some stocks behave strangely.
2. Less is More: In a huge market, the safest "Long-Only" portfolio is often surprisingly small. You don't need to own 500 stocks to be safe; you might only need the 20 "calmest" ones.
3. Solving the Mystery: The authors finally solved a math problem that had been stuck for years regarding how to handle stocks that move in the opposite direction of the market.

In a nutshell: This paper tells us that when building a safe, "no-short-selling" investment portfolio in a world driven by a single economic wave, quality beats quantity. You should focus on a small, carefully selected group of assets rather than trying to hold everything.

Technical Summary

1. Problem Statement

The paper addresses the Long-Only Minimum Variance (LOMV) portfolio optimization problem. Unlike the standard Global Minimum Variance (GMV) portfolio which allows for short selling (negative weights), the LOMV problem imposes non-negativity constraints ($w_i \ge 0$) on asset weights.

• The Challenge: While the unconstrained GMV solution has a closed-form formula ($w_{LS} = \frac{\Sigma^{-1}\mathbf{1}}{\mathbf{1}^\top \Sigma^{-1}\mathbf{1}}$), the long-only solution requires identifying which assets have positive weights (the active set $K$) and which have zero weights. Determining $K$ is generally a combinatorial problem that is computationally difficult and lacks a simple explicit solution, especially when asset betas (factor loadings) can be negative.
• Context: The authors focus on a single-factor (1-factor) covariance model, where the covariance matrix $\Sigma$ is defined as $\Sigma = \sigma^2 \beta \beta^\top + \Delta$. Here, $\beta$ represents asset betas (which can be positive, negative, or zero), $\sigma^2$ is the factor variance, and $\Delta$ is a diagonal matrix of idiosyncratic variances.
• Open Question: Previous rigorous work (e.g., Qi [2021]) solved the LOMV problem for the 1-factor model but only under the restrictive assumption that all betas are positive. The paper aims to resolve the open question of extending this solution to mixed-sign betas.

2. Methodology

The authors employ a combination of convex optimization theory, linear algebra, and high-dimensional probability theory.

A. Finite-Dimensional Analysis (Exact Solution)

1. KKT Conditions: The authors utilize the Karush-Kuhn-Tucker (KKT) conditions for the constrained quadratic optimization problem.
2. Active Set Reduction (Proposition 1): They prove that once the active set $K$ (assets with $w_i > 0$) is known, the optimal weights for these assets are simply the GMV solution calculated on the reduced sub-matrix $\Sigma_K$.
3. Explicit Thresholding (Theorem 1): For the 1-factor model, the authors derive an explicit sequence $\{R_i\}$ based on the ordered betas ($\beta_1 \le \beta_2 \le \dots \le \beta_p$).
   • They define $R_i = \frac{1}{\sigma^2} + \sum_{j=1}^{i-1} \frac{\beta_j}{\delta_j^2}(\beta_j - \beta_i)$.
   • They prove that the sequence $\{R_i\}$ is unimodal (increases then decreases).
   • Key Insight: The active set $K$ consists exactly of the first $\ell$ assets, where $\ell$ is the largest index such that $R_\ell > 0$. This provides a computable $O(\log p)$ method (via bisection) to find the active set without solving a fixed-point equation, even with negative betas.

B. High-Dimensional Asymptotics

The authors analyze the behavior of the active ratio ($k/p$, the proportion of active assets) as the number of assets $p \to \infty$.

• Assumptions: Betas are drawn i.i.d. from a distribution with CDF $F$, and idiosyncratic variances are drawn from an independent distribution.
• Limiting Function: They define a function $G(y) = \int_{-\infty}^y (x^2 - yx) dF(x)$ and identify a critical threshold $\beta^*$ as the root of $G(y)=0$.
• Convergence Analysis: Using the Strong Law of Large Numbers (SLLN) and properties of concave functions, they prove that the empirical active ratio converges to $F(\beta^*)$ (or oscillates around it if $\beta^*$ is an atom).

3. Key Contributions

1. Resolution of the Mixed-Sign Beta Problem: The paper provides the first rigorous, explicit solution for the LOMV portfolio under a 1-factor model where betas can have arbitrary signs (positive, negative, or zero). This resolves the open problem left by Qi [2021].
2. Explicit Characterization of the Active Set: Theorem 1 establishes that the active set is determined by a simple threshold on the ordered betas, identified by the sign change of a computable sequence $R_i$. This avoids the need for iterative fixed-point solvers.
3. Asymptotic Active Ratio Theory: Theorem 2 characterizes the limiting proportion of active assets. It shows that:
   • If betas are non-negative, the active ratio converges to 0 (unless there is an atom at the minimum beta).
   • If negative betas exist, the active ratio converges to $F(\beta^*)$, where $\beta^*$ is determined by the distribution of betas.
4. Convergence Rate for Small Negative Betas: Theorem 3 establishes a quantitative bound for the active ratio when the proportion of non-positive betas is small. Specifically, if the mass of non-positive betas is $\epsilon$, the active ratio scales as $O(\epsilon^{1/3})$. This explains why the active ratio vanishes rapidly as negative betas become rare.

4. Key Results

• Theorem 1 (Explicit Solution): The optimal long-only portfolio includes the first $k$ assets (sorted by beta) where $k = \max \{i : R_i > 0\}$. The weights are given by the standard GMV formula applied to this subset.
• Theorem 2 (Asymptotic Active Ratio):
  • Case 1 (Mixed signs, positive mean): The active ratio converges to $F(\beta^*)$ if $F$ is continuous at $\beta^*$. If $\beta^*$ is an atom, the limit does not exist, and the ratio oscillates between $F(\beta^*-)$ and $F(\beta^*)$.
  • Case 2 (Mixed signs, zero mean): The active ratio converges to 1 (all assets are active).
  • Case 3 (Non-negative betas): The active ratio converges to 0 (if no atom at the minimum beta) or the mass of the atom at the minimum beta.
• Theorem 3 (Continuity at Zero): If the proportion of non-positive betas $F(0) \to 0$, the limiting active ratio $F(\beta^*)$ decays at a rate of $O(F(0)^{1/3})$.
• Numerical Validation: Simulations confirm the theoretical limits and illustrate an "upward bias" for finite $p$ (where the sample active ratio is slightly higher than the asymptotic limit) due to finite-sample fluctuations in the empirical distribution.

5. Significance and Implications

• Practical Portfolio Construction: The paper provides a fast, non-iterative algorithm for constructing LOMV portfolios in single-factor markets, which is highly relevant for large-scale equity portfolios where short selling is restricted.
• Understanding "Sparsity": The results explain a well-known empirical phenomenon: in large universes of assets with mostly positive betas, the long-only minimum variance portfolio is extremely sparse (holds very few assets). The paper quantifies this sparsity, showing it is a direct consequence of the single-factor structure and the rarity of negative betas.
• Theoretical Rigor: By removing the "positive beta" assumption, the paper bridges the gap between theoretical finance models and empirical reality, where negative factor loadings (though rare) do exist.
• Stability Analysis: The identification of non-convergence cases (when $\beta^*$ is an atom) highlights a potential instability in portfolio construction where small changes in the sample composition can lead to large jumps in the number of active assets.

In summary, this paper transforms the LOMV problem from a difficult numerical optimization task into a tractable analytical problem for single-factor markets, providing deep insights into the asymptotic behavior of portfolio sparsity under realistic constraints.