Do Whitepaper Claims Predict Market Behavior? Evidence from Cryptocurrency Factor Analysis
Murad Farzulla, [0.3em] King’s College London Dissensus AI [0.3em] ORCID: 0009-0002-7164-8704 · 2026
Read Do Whitepaper Claims Predict Market Behavior? Evidence from Cryptocurrency Factor Analysis by Murad Farzulla, [0.3em] King’s College London Dissensus AI [0.3em] ORCID: 0009-0002-7164-8704 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-20336.txt.
Introduction
Murad Farzulla asks whether the claims in cryptocurrency whitepapers predict how the coins actually behave in markets. Projects state their value in whitepapers, making claims about function, use cases, and technical ability. The study checks whether those narrative claims line up with observed market behavior, using factor analysis.
Factor analysis groups coins by shared patterns in their returns. If whitepaper claims sort coins the same way the market does, the claims carry real information. If not, investors are reading stories the data do not support. The test gives buyers a check on one of crypto's main sales tools.
Catalog record
| CREATOR | Murad Farzulla, [0.3em] King’s College London Dissensus AI [0.3em] ORCID: 0009-0002-7164-8704 |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-20336
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Do Whitepaper Claims Predict Market Behavior? Evidence from Cryptocurrency Factor Analysis Murad Farzulla, [0.3em] King’s College London Dissensus AI [0.3em] ORCID: 0009-0002-7164-8704 Abstract Cryptocurrency projects articulate their value propositions through whitepapers, making foundational claims about functionality, use cases, and technical capabilities. This study investigates whether these narrative claims align with empirically observed market behavior. We construct a novel pipeline combining natural language processing (NLP) with tensor decomposition to compare three representational spaces: (1) a claims matrix derived from zero-shot classification of 38 whitepapers across 10 semantic categories using BART-MNLI, (2) a market statistics matrix capturing 7 financial metrics for 49 cryptocurrency assets over two years of hourly data (17,543 timestamps), and (3) latent factors extracted via CP tensor decomposition (rank 2, explaining 92.45% of variance). Using Procrustes rotation and Tucker’s congruence coefficient ( ϕ \phi ), we test alignment between narrative and market spaces across 37 common entities. Results indicate weak alignment: claims–statistics ( ϕ = 0.246 \phi=0.246 , p = 0.339 p=0.339 ), claims–factors ( ϕ = 0.058 \phi=0.058 , p = 0.751 p=0.751 ), and statistics–factors ( ϕ = 0.174 \phi=0.174 , p < 0.001 p<0.001 ). Critically, the statistics–factors significance ( p < 0.001 p<0.001 ) validates our methodology: since both matrices derive from the same market data, significant alignment confirms the Procrustes pipeline detects relationships when present, establishing that the narrative null result represents observed weak alignment rather than measurement failure. Inter-model validation using DeBERTa-v3 yields 32% exact agreement (Cohen’s κ = 0.14 \kappa=0.14 ) but 67% relaxed (top-3) agreement, indicating models capture similar semantic neighborhoods with different decision boundaries. Cross-sectional analysis reveals XMR, CRV, YFI, and SOL exhibit positive alignment contributions, while RPL, HBAR, AAVE, and SUSHI show the largest divergence. We interpret these findings as consistent with weak alignment between whitepaper narratives and realized market factor structure. Limited statistical power ( n = 37 n=37 ) precludes distinguishing weak alignment from no alignment, but we can confidently reject strong alignment ( ϕ ≥ 0.70 \phi\geq 0.70 ). Implications for narrative economics, market efficiency, and investment analysis are discussed. Keywords: Cryptocurrency, Tensor Decomposition, NLP, Factor Analysis, Procrustes Rotation, Tucker’s Congruence Coefficient, Zero-Shot Classification JEL Codes: G14, G12, C38, C45 Acknowledgements. The author acknowledges Claude (Anthropic) for assistance with pipeline development, mathematical exposition, and technical writing. All errors, omissions, and interpretive limitations remain the author’s responsibility. Data & Code Availability. Reproducible code and data are available at https://github.com/studiofarzulla/tensor-defi . 1 Introduction Cryptocurrency markets present a unique laboratory for studying the relationship between narrative and price. Unlike traditional equities, where value propositions emerge gradually through earnings reports and analyst coverage, cryptocurrency projects typically articulate comprehensive visions at inception through whitepapers. These foundational documents make explicit claims about functionality, use cases, and technical architecture—claims that should, in principle, relate to how assets behave in markets. The efficient market hypothesis suggests that asset prices reflect available information ( 20 ) . If whitepapers constitute meaningful information about project characteristics, we might expect narrative claims to align with market behavior. Conversely, 52 argues that “narrative economics” drives market dynamics through stories that spread virally, potentially decoupling prices from fundamentals. Recent work has begun examining this in cryptocurrency markets, with 4 documenting significant correlations between prices and sentiment across nearly two thousand cryptocurrencies. Shiller’s framework emphasizes how narratives propagate and influence aggregate behavior; we adapt this insight to test a related but distinct hypothesis about structural alignment between project-level narratives and market factor structure. This tension motivates our central research question: Do whitepaper claims predict market behavior? Specifically, do the functional narratives articulated in project whitepapers align with empirically observed market factor structure? We emphasize that our methodology measures contemporaneous structural alignment between representational spaces rather than predictive relationships; the term “predict” in our research question should be understood as “exhibit structural correspondence with” rather than temporal forecasting. We address this question through a novel methodological pipeline combining natural language processing with tensor decomposition. Our approach constructs three distinct representational spaces: 1. A claims matrix \mathbf{C}\in\mathbb{R}^{N\times K} derived from zero-shot classification of whitepaper text across K=10 semantic categories 2. A market statistics matrix \mathbf{S}\in\mathbb{R}^{M\times J} capturing J=7 financial metrics across M=49 assets 3. Latent factors \mathbf{F}\in\mathbb{R}^{M\times R} extracted from a high-dimensional market tensor via CP decomposition Using Procrustes rotation and Tucker’s congruence coefficient, we then test whether these spaces align—whether assets that make similar claims exhibit similar market behavior. Our findings reveal weak alignment across comparisons ( \phi<0.25 ), with one notable exception: the statistics–factors comparison achieves statistical significance ( p<0.001 ) despite weak magnitude, indicating systematic correspondence between market metrics and tensor-derived factors that narrative claims fail to capture. Specialized tokens (XMR, CRV, YFI, SOL) show positive alignment contributions, while DeFi infrastructure tokens (RPL, HBAR, AAVE, SUSHI) exhibit the largest narrative-market divergence. This result is robust to temporal variation, subsample perturbation, decomposition method (CP vs Tucker), and rank selection. Contributions. This paper makes four contributions: (1) we introduce a reproducible pipeline for comparing textual and market representational spaces in cryptocurrency research, (2) we provide detailed methodological exposition of tensor decomposition for factor extraction, (3) we deliver rigorous empirical evidence on the (mis)alignment between project narratives and market behavior, and (4) we demonstrate that null results in this domain constitute valid findings with implications for narrative economics. The remainder of this paper is organized as follows. Section 2 reviews related work. Section 3 describes our data sources. Section 4 provides detailed methodology with full mathematical exposition. Section 5 presents comprehensive results including sensitivity analyses. Section 6 interprets findings, and Section 7 concludes. 2 Related Work 2.1 Cryptocurrency Narratives and Sentiment Research on cryptocurrency narratives spans social media analysis, whitepaper studies, and sentiment measurement. 17 show that Twitter sentiment predicts Bitcoin returns at short horizons, while 2 find that Elon Musk’s tweets generate significant abnormal returns for mentioned cryptocurrencies. 28 document speculative bubbles driven by narrative contagion across crypto assets. 39 establish that cryptocurrency returns exhibit momentum, size effects, and exposure to market-wide factors distinct from traditional assets. Whitepaper analysis has received growing attention as a signal of project quality. 30 examine ICO whitepaper quality as a signal of project legitimacy, finding that technical depth correlates with fundraising success. 23 show that whitepaper informativeness predicts ICO outcomes. 1 provide the first comprehensive empirical description of ICO determinants, finding that code availability, token presale, and utility design predict success. More recent work has applied sophisticated NLP to whitepaper content: 56 use topic modeling on 5,210 whitepapers to show that technical feature topics predict success while whitepaper informativeness diminishes post-listing. Critically, 45 demonstrates that issuers systematically exaggerate whitepaper claims—exaggeration raises funds short-term but causes token depreciation and platform failure long-term. 50 find that whitepaper length and complexity correlate positively with funds raised, supporting signaling interpretations. 24 show that while expert ratings initially “jam” whitepaper signals, post-listing returns are better predicted by whitepaper content than analyst assessments. However, these studies focus on cross-sectional prediction at issuance rather than ongoing alignment between narrative and market behavior. Indeed, 53 find that clustering cryptocurrencies by whitepaper content does not yield significant differences in time series dynamics, suggesting narratives may matter for initial fundraising but not long-term price behavior. Recent high-profile failures illustrate the gap between whitepaper claims and