The Combinatorial Foundations of Momentum and Factor Structure

The traditional asset-pricing literature has long treated the momentum premium as a market anomaly, a puzzle that demands explanation through risk premia, behavioral biases, or institutional frictions. We challenge this premise at its root. Momentum is not a statistical fluke. It is a mathematical theorem. By reframing equity returns as ordered sequences of real numbers, we demonstrate that momentum-type structure is model-free and assumption-free, emerging necessarily from the mathematics of sequences themselves.

THE MATHEMATICAL FOUNDATION
‍Our foundation begins with the Pigeonhole Principle, which establishes that repetition is unavoidable when distributing objects among fewer categories than objects. When extended to a single stock via the Erdős–Szekeres Theorem of 1935, this logic proves that any return series of sufficient length must contain a monotone subsequence of guaranteed minimum length.
In a standard trading year of 252 days, the theorem delivers a precise and unconditional result:

• A momentum or mean-reversion run of at least 16 consecutive days is mathematically guaranteed
• This guarantee holds regardless of market efficiency, investor rationality, or economic regime
• No asset-pricing model, risk factor, or behavioral assumption is required or invoked

"Momentum does not need to be explained. It needs to be expected. The Erdős–Szekeres Theorem makes its existence as certain as arithmetic."

CROSS-SECTIONAL STRUCTURE
‍The structural inevitability of momentum does not stop at the single-stock level. Through Ramsey Theory, we show that in any sufficiently large equity universe, the pairwise return-correlation graph must contain a monochromatic clique, a subset of stocks moving together as a coherent factor cluster.
The implication is significant. Factor structures are not statistical artifacts uncovered after the fact by Principal Component Analysis. They are combinatorial necessities, guaranteed to exist before a single data point is examined.
Dilworth's Theorem sharpens this further, revealing that momentum and mean-reversion are not opposing market forces but mathematical duals of the same object:

• The depth of any momentum structure exactly determines the minimum number of mean-reversion segments required to partition the full return series
• Neither phenomenon is primary. Each implies and defines the other

DURATION AND SIGNIFICANCE TESTING
To move from existence to measurement, we establish a baseline under a structureless null hypothesis. The expected length of the longest momentum run scales as 2 times the square root of time, a result due to Logan, Shepp, Vershik, and Kerov. The fluctuations around this baseline follow the Tracy-Widom distribution, a result from random matrix theory, enabling a rigorous significance test to distinguish genuine momentum regimes from statistical noise.
Practical consequences for portfolio construction:

• Doubling the lookback window increases expected signal duration by only a factor of 1.41, not 2
• Lookback selection is therefore not arbitrary. It has a mathematically principled foundation
• Signals that exceed Tracy-Widom bounds can be classified as statistically significant momentum regimes with formal confidence

THREE-LAYER DECOMPOSITION
We propose organizing the problem into three distinct and non-overlapping layers, each with its own tools, uncertainty profile, and research agenda.

Layer 1 : Existence

• Unconditional guarantee, requiring no data and no assumed model
• Grounded in Erdős–Szekeres and Ramsey Theory
• Momentum and factor structure must exist in any return series of sufficient length

Layer 2 : Detection

• Conditional on Layer 1
• The challenge is identifying these structures in real time
• Tools include Hidden Markov Models, Kalman Filters, and regime-switching frameworks
• Irreducible uncertainty is inherent to this layer and should be modeled, not eliminated

Layer 3 : Monetization

• Conditional on Layers 1 and 2
• Profitability depends on transaction costs, market impact, signal decay, and crowding dynamics
• This is where mathematics meets markets, and where most alpha erodes

‍CONCLUSION
The existence of momentum is not an open empirical question. It is a closed mathematical one. The open questions, and the ones that matter for institutional investors, sit entirely within Layers 2 and 3: how to detect these structures with precision, and how to extract value from them before the market does. The anomaly was never in the market. It was in how the question was framed.

For institutional use only. This material is provided for informational purposes and does not constitute investment advice.© 2026 LevUp. All rights reserved.

‍

‍

‍