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

arXiv:2604.09986 · q-fin.MF · Submitted 2026-04-11 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: I'm Ruth.

Warren: And I'm Warren, here with Iris.

Ruth: We're looking at the paper: "The Long-Only Minimum Variance Portfolio in a One-Factor Market".

Warren: Here's the gist of the paper. The long-only minimum variance (LOMV) portfolio in a one-factor market is studied under a covariance model with arbitrary sign asset betas to provide an explicit solution and characterization of active assets, resolving an open question regarding mixed-sign betas <ref:2604.09986#pg12>.

Ruth: Explicit Solution for the LOMV Problem The authors provide a new explicit and easily computable solution (Theorem 1) for the LOMV problem in terms of the parameters σ 2, β i, δ i, along with a rigorous proof that applies to a general 1-factor model without requiring that the betas are all positive.

Warren: The solution is characterized by determining the active set K through a threshold on the ordered betas identified directly from the sign changes of an explicit sequence, without needing to solve a fixed-point equation.

Ruth: This approach resolves an open question concerning the extension to mixed-sign betas.

Paper summary: Ruth: So this paper, "The Long-Only Minimum Variance Portfolio in a One-Factor Market: Theory and Asymptotics," it's looking at how to find that minimum risk portfolio when you have to only go long positions. It sets up this scenario with just one common factor driving all the assets.

Warren: Exactly, Ruth. The main thing they're tackling is the long-only minimum variance problem, which is different from the standard long-short setup because you can't take short positions on any asset in that portfolio.

Iris: And what they claim upfront is that they provide an explicit solution for this LOMV problem in terms of a few parameters like sigma squared, the betas, and some other things called delta i. They also give a rigorous way to characterize which assets are active—meaning which ones actually get a positive weight in the portfolio.

Ruth: That's interesting because they say this new approach resolves an open question about extending these results to situations where the asset betas can be both positive and negative, not just all positive betas.

Warren: Right, so they aren't stuck in that old problem anymore. They give a specific way to find the active set by looking at sign changes in an explicit sequence, instead of having to solve some kind of fixed-point equation.

Iris: That characterization is important because it tells us exactly when an asset gets included or excluded from the long-only portfolio based on its beta value compared to some threshold.

Ruth: And then they look at what happens when you have a really big number of assets, where we talk about the dimension p going to infinity, and how that affects things.

Warren: In that high-dimensional regime, they prove that the proportion of active assets in this LOMV portfolio converges in almost all cases to something determined by some function F of those betas.

Iris: And that function F depends on a root, beta star, which comes from an integral equation they set up based on the distribution of those betas. It connects the structure of the asset returns to how many assets end up in the portfolio.

Paper summary: Ruth: That convergence is interesting because when you have a continuous distribution for those betas and all of them are positive, meaning F(zero) equals zero, then this active ratio converges to zero <ref:2604.09986#pg1,active ratio converges to zero>.

Warren: That makes sense, it suggests that if there are no negative factor loadings in the assets you're holding, you end up with almost no assets actually getting weight in the LOMV portfolio.

Iris: But they also discuss what happens when F(zero) is small but greater than zero, and under certain conditions about moments and concentration bounds, they show that the active ratio converges at a rate related to F(zero) raised to the power of one over three <ref:2604.09986#pg1>.

Ruth: That convergence rate is what's really important for understanding how quickly this portfolio settles down as you add more assets. It shows that if there are few negative betas, the portfolio behaves in a predictable way.

Warren: But there are some exceptions they found where things don't converge nicely, specifically when beta star is an atom of the distribution F. That causes the active ratio to oscillate between two different values, F beta star minus and F beta star.

Iris: That oscillation happens because of how randomly the remaining assets are composed, not just because of some specific variance issues. It's driven by that random composition.

Ruth: So, even though they have these convergence results and rates, there are still these non-convergence phenomena depending on the exact nature of the beta distribution F.

Warren: When we look at the proof for Theorem three they show that in specific cases where negative betas are rare, like when F(zero) is small, this oscillation doesn't happen as badly as you might think <ref:2604.09986#pg2>.

Iris: They prove that the mass of the distribution function F on a certain interval around beta star is bounded by something related to F(zero) raised to the power of one over three <ref:2604.09986#pg1>. This gives us a concrete bound on how fast it converges in those tricky situations.

Paper summary: Ruth: That bound is significant because it explains why, empirically, we often see the long-only constraint binding on most available assets when negative market factor loadings are rare in our real data.

Warren: It links the theoretical limit back to what we actually observe when p gets large in simulations. Their simulations confirm that as the number of assets p increases, the active ratio does indeed approach that theoretical limit F beta star.

Iris: The simulation also showed how this limit depends on that proportion of non-positive betas, which is consistent with the convergence rate result from Theorem three when F(zero) is small <ref:2604.09986#pg2>.

Ruth: And they pointed out a slight upward bias in the active ratio observed for moderate numbers of assets, but they explained that's due to a term involving n squared over sigma squared, which vanishes as p goes to infinity.

Warren: So, the numerical evidence aligns with their theoretical predictions regarding the limit and how it depends on those distribution properties. They also showed how the convergence rate behaves in different scenarios.

Iris: And they clearly separated Case one where there's oscillation between F beta star minus and F beta star, from Case three where the limit is stable and equals F beta star almost surely <ref:2604.09986#pg2>. That distinction is key for understanding the behavior of these portfolios.

Ruth: It seems that this paper gives us a solid framework to analyze how long-only constraints interact with factor models when those models allow for mixed-sign betas.

Warren: Yeah, it offers a way to explicitly characterize the active set and understand the asymptotic behavior of the portfolio proportion, which is helpful for thinking about how these constraints play out in practice.

Iris: It’s important to note that they do have limitations on what they can prove universally, especially when dealing with those non-convergence scenarios driven by atoms in the distribution F.

Ruth: So, overall, the paper presents a detailed theoretical treatment of the long-only minimum variance portfolio under a one-factor model, giving us both an explicit solution and some very clear insights into its behavior in high dimensions.

Conclusion: Ruth: So, to wrap up, this paper is about figuring out exactly how to construct the long-only minimum variance portfolio when you only have one factor driving everything and you can't short anything.

Warren: Yeah, they give us a concrete mathematical way to find that portfolio and then tell us which assets actually get included in it.

Iris: And they do this without needing those complicated fixed-point equations we usually have to solve. They find the active set just by looking at sign changes in an explicit sequence of numbers derived from the betas.

Ruth: It’s a big deal because before, extending these types of solutions to models where some asset betas are negative was a real open question.

Warren: Right, so they solved that problem by giving us this new explicit formula that works even when you have mixed signs in those factor loadings.

Iris: The asymptotic results are also pretty telling, showing how the proportion of active assets behaves when you add more and more assets to your model.

Ruth: They find that under certain conditions on the distribution of those betas, this active ratio settles down to a specific value determined by some root beta star.

Warren: And they show that if there are few negative betas in your universe, this limit is small, which makes sense for real-world data where you don't have tons of assets with negative factor loadings.

Iris: But they also flag this oscillation phenomenon when the distribution has a specific kind of point called an atom, which means things don't settle down perfectly in those cases.

Ruth: It suggests that while the theory gives us strong limits, you still have to be careful about how the underlying probability distribution behaves for these small numbers of negative betas.

Warren: So what does this mean for us in practice? It means we can finally build a long-only strategy with a much clearer picture of which assets are doing the work and how many there will be.

Florida State University

q-fin.MF

Submitted: 2026-04-11

Updated: 2026-11-06

Comments: 29 pages, 3 figures

License: http://creativecommons.org/licenses/by-sa/4.0/

Importance score: 80/100

The gist: The long-only minimum variance (LOMV) portfolio in a one-factor market is studied under a covariance model with arbitrary sign asset betas to provide an explicit solution and characterization of…

Key concepts

Long-Only Minimum Variance (LOMV) Portfolio
This is a portfolio strategy designed to minimize risk while only holding assets with positive weights. The study finds an explicit formula for this portfolio's weight allocation, which is crucial for understanding optimal long-only investment decisions in a single-factor market.
Active Set K
The active set K represents the subset of assets that will have positive weights in the optimal LOMV portfolio. The paper defines how to determine this set by checking an inequality involving ordered betas, which helps identify which assets should be included in the long-only portfolio.
Active Ratio Convergence
This concept describes how the proportion of active assets in a large portfolio behaves as the number of assets increases. The paper proves that this ratio converges to a value determined by the distribution of asset betas, specifically F(beta*), under certain conditions.

Terminology

Summary

The long-only minimum variance (LOMV) portfolio in a one-factor market is studied under a covariance model with arbitrary sign asset betas to provide an explicit solution and characterization of active assets, resolving an open question regarding mixed-sign betas <ref:2604. p. 0

Where the paper says it

edu This draft: 6 April 2026 Abstract We study the long-only minimum variance (LOMV) portfolio under a one-factor covariance model with asset betas of arbitrary sign. We provide an explicit solution in terms of the set of active (positive weight) assets, and provide an explicit and computable characterization of the active set.

09986#pg12>.

Explicit Solution for the LOMV Problem

The authors provide a new explicit and easily computable solution (Theorem 1) for the LOMV problem in terms of the parameters σ 2, β i, δ i, along with a rigorous proof that applies to a general 1-factor model without requiring that the betas are all positive. The solution is characterized by determining the active set K through a threshold on the ordered betas identified directly from the sign changes of an explicit sequence, without needing to solve a fixed-point equation. p. 4

Where the paper says it

The Long-Only Minimum Variance Portfolio in a 1-Factor Market to extend the solution to the case when betas can have mixed signs. The present paper resolves this open question: Theorem 1 provides a complete, rigorous solution for the LOMV problem under the 1-factor model with betas of arbitrary sign, showing that the active set K is still determined by a single threshold on the ordered betas, identified directly from the sign changes of an explicit sequence {Ri}, without the need to solve a fixed point equation.

This approach resolves an open question concerning the extension to mixed-sign betas. p. 0
Where the paper says it

We provide an explicit solution in terms of the set of active (positive weight) assets, and provide an explicit and computable characterization of the active set. As a corollary we resolve an open question of Qi [2021] concerning the extension to mixed-sign betas.

Asymptotic Behavior of Active Assets

In the high-dimensional regime p → ∞ where the betas are drawn from a distribution with cdf F, it is proven that the proportion of active assets (the active ratio) in the LOMV portfolio converges in almost all cases to F(β∗), where β∗ ≥ 0 is the root of an explicit integral equation determined by F. When F is continuous and all betas are positive (F(0) = 0), the active ratio converges to zero. p. 0

Where the paper says it

This is a variation of a result first appearing in Bernstein and Shkolnik [2025]. In particular, when F is continuous and all betas are positive (F(0) = 0), the active ratio converges to zero.

Under mild moment conditions and concentration bounds, the convergence rate is established as F(β∗) = O(F(0)1/3) as F(0) → 0. p. 0
Where the paper says it

In particular, when F is continuous and all betas are positive (F(0) = 0), the active ratio converges to zero. When F(0) > 0 is small, under mild moment conditions and concentration bounds we establish the convergence rate F(β ∗ ) = O(F(0)1/3 ) as F(0) → 0.

Characterization of Active Set Dynamics

The active set K is determined by the condition that for any asset i, its weight wL i > 0 if and only if a certain inequality involving the ordered betas holds. This leads to Corollary 2, which states that either K = P and the long-short solution is already long-only, or else K ≠ P, with wi > 0 if and only if βi < 1σ squared + P j<β 2jδ 2jP jβjδ 2j. p. 7

Where the paper says it

(19) The active set described this way is a consequence of the following corollary of the proof of Theorem 1: Corollary 2. Either K = P and the long-short solution w LS is already long-only (w LS = w L), or else K ̸= P, Pk j=1 βj δ 2 j > 0, and wi > 0 if and only if βi < 1 σ2 + Pk j=1 β 2 j δ 2 j Pk j=1 βj δ 2 j .

Convergence Analysis in High Dimensions

Theorem 3 establishes that under specific conditions on the distribution F, the limiting active ratio F(β∗) is small if the proportion F(0) of non-positive betas is small. Specifically, it shows that F(β∗) = O(F(0)1/3), with the constant depending on distributional bounds. p. 9

Where the paper says it

The next theorem shows, under reasonable assumptions, that the limiting active ratio F(β ∗ ) is small if the proportion F(0) of non-positive betas is small. More specifically, F(β ∗ ) = O(F(0)1/3 ), with the constant depending on distributional bounds.

This result is significant because it explains why the long-only constraint binds on most available assets when negative market factor loadings are rare empirically. p. 3
Where the paper says it

Our results serve as a rigorous quantitative justification for this conclusion. Since negative market factor loadings are rare empirically, Theorems 2 and 3 explain why the long-only constraint binds on most of the available assets.

Non-Convergence Phenomena

The analysis of Case 1 of Theorem 2 illustrates a non-convergence phenomenon where the limit does not exist if β∗ is an atom of F, causing the active ratio to oscillate between F(β∗−) and F(β∗). This oscillation is driven by the random composition of remaining assets rather than just idiosyncratic variances. p. 2

Where the paper says it

The Long-Only Minimum Variance Portfolio in a 1-Factor Market a 1-factor (“single index”) model, in which the covariance matrix takes the form Σ = σ 2ββ⊤ + ∆, (5) where σ 2 > 0 is the variance of the single factor return, β is a p-vector of exposures of the asset returns to the factor, and ∆ is a diagonal matrix of variances δ 2 i > 0, i ∈ P, of asset specific returns uncorrelated with the common factor return. Figure 1: Example comparison of long-short (blue) and long-only (black) minimum variance portfolio weights plotted against beta for p = 5000 assets under the single-factor model (5) with βi drawn iid from a Normal N (1, 0.

Numerical Confirmation

Numerical simulations confirm that as the number of assets p increases, the active ratio kp/p approaches its theoretical limit F(β∗). Furthermore, the simulation demonstrates how this limit depends on the proportion of non-positive betas, consistent with Theorem 3. p. 9

Where the paper says it

In practice one may find the proportion P(β ≤ 0) = F(0) is small but positive, and any negative betas tend to be small in magnitude. The next theorem shows, under reasonable assumptions, that the limiting active ratio F(β ∗ ) is small if the proportion F(0) of non-positive betas is small.

The upward bias observed for moderate p is explained by a term ν 2/(pσ 2) which vanishes as p → ∞. p. 9
Where the paper says it

The Long-Only Minimum Variance Portfolio in a 1-Factor Market In the non-negative beta case P(β < 0) = 0 treated by [Clarke et al. , 2011] and [Qi, 2021], in the realistic case where F has no atoms, Theorem 2 tells us to expect that the active ratio asymptotically vanishes as p → ∞.

Proof of Theorem 3

The proof of Theorem 3 involves showing that the mass Fn places on (0, y∗ n] is O(Fn(0)1/3) by combining bounds derived from concentration function conditions. This leads to the conclusion that Fn(y∗ n) = O(Fn(0)1/3), which proves the convergence of the active ratio to zero. p. 0

Where the paper says it

In particular, when F is continuous and all betas are positive (F(0) = 0), the active ratio converges to zero. When F(0) > 0 is small, under mild moment conditions and concentration bounds we establish the convergence rate F(β ∗ ) = O(F(0)1/3 ) as F(0) → 0.

Proof of Theorem 1

The proof of Theorem 1 is established by analyzing the monotonicity properties of the sequence R i, showing that k = max i: R i > 0 and K = P up to a threshold index. This completes the determination of the long-only minimum risk portfolio as soon as K is known. p. 5

Where the paper says it

(8) This is the unique solution of the k-dimensional long-short problem (1) with Σ replaced by Σ K, and w L can be recovered from w K by adding back zero entries for the deleted assets. Proposition 1 allows us to determine the long-only minimum risk portfolio as soon as we know the set K of active assets.

The final solution wL is then recovered using Proposition 1. p. 5
Where the paper says it

(8) This is the unique solution of the k-dimensional long-short problem (1) with Σ replaced by Σ K, and w L can be recovered from w K by adding back zero entries for the deleted assets. Proposition 1 allows us to determine the long-only minimum risk portfolio as soon as we know the set K of active assets.

Proof of Theorem 2

The proof of Theorem 2 treats the three cases based on the distribution F, showing that in Case 3 (non-negative betas), the limit exists and equals F(β∗) = P(β = β∗). In Case 1, where P(β 0 and E[β] > 0, the limit does not exist because the mass at the atom β∗ causes oscillation between F(β∗−) and F(β∗). p. 8

Where the paper says it

) If P(β < 0) = 0, then β ∗ = inf{x ≥ 0 : F(x) > 0} is the left endpoint of the support of F, and lim p→∞ kp/p = F(β ∗ ) = P(β = β ∗ ) a. In particular, the limit is zero whenever F has no atom at β ∗ , and equals the atom mass otherwise.

In Case 2, where E[β] = 0, the active ratio converges to 1. p. 0
Where the paper says it

As a corollary we resolve an open question of Qi [2021] concerning the extension to mixed-sign betas. In the high-dimensional regime p → ∞ where the betas are drawn from a distribution with cdf F, we prove that the proportion of active assets (the active ratio) in the LOMV portfolio converges in almost all cases to F(β ∗ ), where β ∗ ≥ 0 is the root of an explicit integral equation determined by F.

The proof of Theorem 3 relies on Lemma 8 establishing an upper bound on Rn as Rn ≤ (K + D√K)εn. This bound, combined with Lemma 9 providing a lower bound for Ln, leads to the final result that Fn(y∗ n) = O(Fn(0)1/3), proving the convergence rate. p. 3

Where the paper says it

We show that in almost all cases, lim p→∞ kp/p = F(β ∗ ), where β ∗ ≥ 0 is the root of an explicit integral equation determined by F, but there are some cases where the limit does not exist. With some moment and concentration bounds on the distribution of betas, we also show that the asymptotic active ratio tends to zero as the fraction of non-positive betas tends to zero (Theorem 3), and provide a cube-root quantitative rate bound for the convergence.

Proof of Proposition 1

Proposition 1 is proved by showing that the solution w = wL satisfies the Karush-Kuhn-Tucker (KKT) conditions. The subsequent derivation using the Woodbury identity and Lemma 1 establishes that K = P and k = p, leading to the explicit form of wL. p. 5

Where the paper says it

It remains to determine K, and for the single-factor case this can be accomplished by an explicit method described next in Theorem 1. We henceforth restrict to the case of a single factor covariance matrix Σ of the form Σ = σ 2ββ⊤ + ∆, (9) as discussed above.

The proof also confirms that CK ≥ 0. p. 2
Where the paper says it

This is the context for the present article. The contributions of this article in this general 1-factor setting may be summarized as follows: • We provide a new explicit and easily computable solution (Theorem 1) for the LOMV problem in terms of the parameters σ 2 , βi , δi , along with a rigorous proof that applies to a general 1- factor model without requiring that the betas are all positive.

Proof of Theorem 1 Part 1

Part 1 of Theorem 1 follows solely from the monotonicity of the sequence R i, showing that k = l and K = i for i=1 to k. This is achieved by verifying that R k > 0 and R k+1 ≤ 0, which implies k = max i: Ri > 0. p. 6

Where the paper says it

The Long-Only Minimum Variance Portfolio in a 1-Factor Market In particular, if ℓ = max i : Ri > 0 , (13) then i ≤ ℓ if and only if Ri > 0, and the sequence {Ri} crosses zero at most once. Let w L denote the LOMV solution of problem (2), K = {i ∈ P : w L i > 0} denote the set of active assets, and let k = |K| denote the number of active assets.

Proof of Theorem 1 Part 2

Part 2 of Theorem 1 determines K by showing that i ∈ K if and only if BK > βiCK, which is established by Lemma 1. The final conclusion is that wL i = wK i for i=1 to k and wL i = 0 for i=k+1 to p. p. 3

Where the paper says it

This suggests that the long-only constraints bind significantly when present. Our results serve as a rigorous quantitative justification for this conclusion.

Proof of Theorem 2 Case 3

In Case 3, where P(β < 0) = 0, the proof shows that G has a strict sign change at β∗, leading to the conclusion that lim inf kp/p = lim sup kp/p = F(β∗). The mass m at β∗ is included in the active set, so kp/p → m a. p. 8

Where the paper says it

) If P(β < 0) = 0, then β ∗ = inf{x ≥ 0 : F(x) > 0} is the left endpoint of the support of F, and lim p→∞ kp/p = F(β ∗ ) = P(β = β ∗ ) a. In particular, the limit is zero whenever F has no atom at β ∗ , and equals the atom mass otherwise.

s.

Proof of Theorem 3 Case 3

In Case 3, the lower bound lim inf kp/p ≥ F(β∗) is shown by demonstrating that Hp > 0 for all p, thus kp/p → m a.s. The upper bound confirms lim sup kp/p ≤ F(β∗) a. p. 3

Where the paper says it

We show that in almost all cases, lim p→∞ kp/p = F(β ∗ ), where β ∗ ≥ 0 is the root of an explicit integral equation determined by F, but there are some cases where the limit does not exist. With some moment and concentration bounds on the distribution of betas, we also show that the asymptotic active ratio tends to zero as the fraction of non-positive betas tends to zero (Theorem 3), and provide a cube-root quantitative rate bound for the convergence.

p. 3
Where the paper says it

We show that in almost all cases, lim p→∞ kp/p = F(β ∗ ), where β ∗ ≥ 0 is the root of an explicit integral equation determined by F, but there are some cases where the limit does not exist. With some moment and concentration bounds on the distribution of betas, we also show that the asymptotic active ratio tends to zero as the fraction of non-positive betas tends to zero (Theorem 3), and provide a cube-root quantitative rate bound for the convergence.

p. 3
Where the paper says it

We show that in almost all cases, lim p→∞ kp/p = F(β ∗ ), where β ∗ ≥ 0 is the root of an explicit integral equation determined by F, but there are some cases where the limit does not exist. With some moment and concentration bounds on the distribution of betas, we also show that the asymptotic active ratio tends to zero as the fraction of non-positive betas tends to zero (Theorem 3), and provide a cube-root quantitative rate bound for the convergence.

p. 3
Where the paper says it

We show that in almost all cases, lim p→∞ kp/p = F(β ∗ ), where β ∗ ≥ 0 is the root of an explicit integral equation determined by F, but there are some cases where the limit does not exist. With some moment and concentration bounds on the distribution of betas, we also show that the asymptotic active ratio tends to zero as the fraction of non-positive betas tends to zero (Theorem 3), and provide a cube-root quantitative rate bound for the convergence.

s., thus the limit exists and equals F(β∗) = m a. p. 8
Where the paper says it

In particular, the limit is zero whenever F has no atom at β ∗ , and equals the atom mass otherwise. In Cases 2 and 3, the limit lim kp/p always exists.

p. 8
Where the paper says it

In particular, the limit is zero whenever F has no atom at β ∗ , and equals the atom mass otherwise. In Cases 2 and 3, the limit lim kp/p always exists.

p. 8
Where the paper says it

In particular, the limit is zero whenever F has no atom at β ∗ , and equals the atom mass otherwise. In Cases 2 and 3, the limit lim kp/p always exists.

p. 8
Where the paper says it

In particular, the limit is zero whenever F has no atom at β ∗ , and equals the atom mass otherwise. In Cases 2 and 3, the limit lim kp/p always exists.

s.

Proof of Theorem 3 Case 1

In Case 1, where P(β 0 and E[β] > 0, the oscillation between F(β∗−) and F(β∗) is shown by analyzing the sign of Hp, leading to lim inf kp/p = F(β∗−) a.s. and lim sup kp/p = F(β∗) a. p. 8

Where the paper says it

(Negative betas, positive mean. ) If P(β < 0) > 0 and E[β] > 0, then G has a unique zero β ∗ > 0, and lim inf p→∞ kp/p = F(β ∗−) > 0, lim sup p→∞ kp/p = F(β ∗ ).

s. The oscillation is driven by the random composition of the remaining assets. p. 8
Where the paper says it

The ideas here can also be adapted to the case where the betas and deltas are correlated, which we omit. Adapting the notation of previous sections, let kp ≤ p denote the number of active assets in the LOMV solution in dimension p, so that the remaining p − kp weights are zero.

p. 8
Where the paper says it

The ideas here can also be adapted to the case where the betas and deltas are correlated, which we omit. Adapting the notation of previous sections, let kp ≤ p denote the number of active assets in the LOMV solution in dimension p, so that the remaining p − kp weights are zero.

Proof of Theorem 3 Case 2

In Case 2, where P(β 0 and E[β] = 0, G has no zero on [0, ∞)[so β∗(p) → ∞ a.s., implying kp/p → 1 a.s.

In Case 3, where P(β 0 for all p, thus kp/p → m a.s. The upper bound confirms lim sup kp/p ≤ F(β∗) a.s., thus the limit exists and equals F(β∗) = m a.s.

Proof of Theorem 3 Case 1 Rate

The convergence rate in Case 1 is established by showing that qn ≤ 27 (K + D√K) M squared 1/3 εn 1/3, which implies Fn(y∗ n) = O(Fn(0)1/3), proving the cube-root rate of convergence. This result is derived from Lemma 8 and Lemma 9. p. 0

Where the paper says it

In the high-dimensional regime p → ∞ where the betas are drawn from a distribution with cdf F, we prove that the proportion of active assets (the active ratio) in the LOMV portfolio converges in almost all cases to F(β ∗ ), where β ∗ ≥ 0 is the root of an explicit integral equation determined by F. This is a variation of a result first appearing in Bernstein and Shkolnik [2025].

Proof of Theorem 3 Case 3 Rate

In Case 3, the convergence rate is established by showing that Gp(β∗) > 0 for all p, leading to kp/p → m a.s. The upper bound confirms lim sup kp/p ≤ F(β∗) a.s., thus the limit exists and equals F(β∗) = m a.s.

Proof of Theorem 3 Case 1 Oscillation

The oscillation in Case 1 is shown by analyzing the sign of Hp, which changes sign infinitely often a.s., leading to lim inf kp/p = F(β∗−) and lim sup kp/p = F(β∗). p. 8

Where the paper says it

(Negative betas, positive mean. ) If P(β < 0) > 0 and E[β] > 0, then G has a unique zero β ∗ > 0, and lim inf p→∞ kp/p = F(β ∗−) > 0, lim sup p→∞ kp/p = F(β ∗ ).

This oscillation is driven by the random composition of the remaining assets. p. 8
Where the paper says it

The ideas here can also be adapted to the case where the betas and deltas are correlated, which we omit. Adapting the notation of previous sections, let kp ≤ p denote the number of active assets in the LOMV solution in dimension p, so that the remaining p − kp weights are zero.

In Case 2, where E[β] = 0, G has no zero on [0, ∞)[so β∗(p) → ∞ a.s., implying kp/p → 1 a.s.

Proof of Theorem 3 Case 3 Limit

In Case 3, the lower bound lim inf kp/p ≥ F(β∗) is shown by demonstrating that Hp > 0 for all p, thus kp/p → m a.s. The upper bound confirms lim sup kp/p ≤ F(β∗) a.s., thus the limit exists and equals F(β∗) = m a.s.

Proof of Theorem 3 Case 1 Limit

In Case 1, where P(β 0 and E[β] > 0, the limit does not exist because the mass at the atom β∗ causes oscillation between F(β∗−) and F(β∗). The oscillation is driven by the random composition of the remaining assets.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Explicit active set determination for long-only portfolios: The improved system can determine which assets are active in a long-only minimum variance portfolio by identifying the set K using a threshold on the ordered betas, eliminating the need to solve a fixed point equation.

  2. Asymptotic active ratio prediction: The system can predict the proportion of active assets in high-dimensional regimes by calculating F(beta star), where beta star is determined by an explicit integral equation derived from the distribution F of asset betas.

  3. Convergence rate analysis for sparse markets: For cases where non-positive betas are rare, the system can quantify the convergence rate of the active ratio using a cube root bound, showing that the asymptotic active ratio tends to zero as the fraction of non-positive betas tends to zero.

  4. Handling mixed-sign betas: The improved system resolves an open question by providing a complete solution for long-only portfolios under a one-factor model with arbitrary sign betas, moving beyond prior restrictions.

  5. Robustness assessment against distribution uncertainty: The system can assess the stability of the active set determination by showing that the active set K and the weights of the active assets in wL are unaffected when adding new assets whose betas are greater than or equal to beta k plus one.

Abstract

We study the long-only minimum variance (LOMV) portfolio, an object of interest going back to Markowitz, and the simplest non-trivial instance of quantitative portfolio construction. Despite its importance, aspects of the LOMV remain mysterious. Here we examine the base case of a general one-factor covariance model, and provide an explicit characterization of the solution. As a corollary we resolve an open question of Qi (2021) for this setting. In the high-dimensional regime of many assets, we prove that the active ratio (proportion of assets with non-zero weight in the LOMV portfolio) converges in almost all cases to a limit expressed in terms of the distribution of asset betas. In particular, when asset betas are all positive and drawn from a continuous distribution, the active ratio tends to zero in the high dimensional limit, and close to zero when the fraction of negative betas is small.

Related papers