{"id":"090fac83-6cc3-4f55-8a95-d9c6164998a7","arxiv_id":"2606.25466","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.5,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Weekly Granger-causality networks of 1-minute crypto returns show Ethereum dominating out-strength while Bitcoin declines and top ranks turn over across 17 assets from 2020–2025.","lead":"High-frequency crypto returns from 2020–2025 form directed Granger networks whose out-strength rankings show Ethereum as the persistent top influencer while Bitcoin declines and the top-five set turns over heavily. The result maps a competitive, non-stable influence hierarchy rather than a fixed set of dominant coins.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"Normalized out-strength rankings across a growing asset universe (N~30\to390) may mix genuine influence dynamics with statistical artifacts of changing FDR power and network density.","rationale":"The reader correctly flags the linear-GC-as-influence assumption as the weakest link and already assigns CONDITIONAL. The concern above is distinct but complementary: even if every weekly GC network is a faithful snapshot, the snapshots are not guaranteed to be commensurable once N and the FDR threshold change dramatically. Because the paper’s strongest empirical claim is precisely the multi-year evolution of those rankings, temporal comparability is load-bearing. The proposed fixed-universe recomputation is a direct, self-contained check that requires no new data sources. It does not overturn the existing CONDITIONAL verdict; it merely sharpens one concrete reason why the claim remains provisional until the check is performed. No stronger internal inconsistency was found; the stationarity filter, FDR procedure, and PC1 control are carefully executed.","tokens_in":12375,"tokens_out":571,"duration_ms":36424,"concrete_test":"Restrict the entire analysis to the fixed subset of ~30 cryptocurrencies that trade continuously from Jan 2020 through Mar 2025; recompute all 275 weekly GC networks, FDR-corrected links, out-strengths, and quarterly rankings on this constant node set only. If ETH remains #1, BTC still declines monotonically, and top-5 Jaccard turnover stays high, the dynamics are robust; if the hierarchy stabilizes or the BTC decline disappears, the published non-stability is driven by N-growth and FDR effects.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (ETH persistently #1, BTC gradual decline, 17 assets cycling through top-5, non-stable competitive hierarchy) treats quarterly-averaged out-strength (normalized by contemporaneous N) as a temporally comparable influence score. Methods §3 apply Benjamini–Hochberg FDR at fixed q=0.05 to n=N(N-1) tests whose number grows by two orders of magnitude; detection thresholds therefore tighten while the set of candidate nodes expands. The subsequent /N normalization (Results, Fig. 5 caption) assumes that any density change is uniform and that new entrants do not systematically alter the relative out-strength of incumbents. If either assumption fails, the reported BTC decline and elevated top-5 turnover after 2022 partly reflect the expanding universe and stricter multiple-testing correction rather than a pure reorganization of causal influence among a stable core. The PC1-robustness check (Fig. 6) and volume covariate (Eq. 6) do not address this temporal-comparability issue.","agreement_with_reader":"partial"},"referee_report":{"model":"grok-4.5","summary":"The manuscript constructs weekly directed, weighted networks of cryptocurrencies from statistically significant Granger-causal links among 1-minute log-returns (Kraken, Jan 2020–Mar 2025, 275 windows). After ADF stationarity filtering and Benjamini–Hochberg FDR control (q=0.05), link weights are the log residual-variance ratios of restricted versus full VARs (BIC lag selection). The authors report heavy-tailed normalized returns, heavy-tailed weight and strength distributions, a sublinear scaling sout∼sin^α (α≈0.91) that survives a volume covariate, and a quarterly ranking by N-normalized out-strength in which Ethereum remains #1, Bitcoin declines, and 17 distinct assets occupy the top five, interpreted as a competitive, non-stable hierarchy of influence. A PC1-removal robustness check leaves out-strengths and weights essentially unchanged (Spearman ρ≈0.999/0.995).","tokens_in":12643,"tokens_out":951,"duration_ms":8972,"significance":"If the temporal rankings are comparable, the paper supplies a longer-horizon, higher-resolution extension of earlier GC-network studies of crypto (notably Scagliarini et al.), documenting a clear shift of dominance from Bitcoin toward Ethereum and substantial top-five turnover. Strengths include systematic stationarity testing, explicit FDR control, BIC lag selection, a volume-controlled regression for the sublinear scaling, and a transparent common-factor robustness check. These elements make the work a useful empirical contribution to the network analysis of high-frequency crypto markets, provided the comparability of the ranking measure across a growing asset universe is secured.","major_comments":[{"comment":"Results §4 / Fig. 5: the central ranking claim rests on quarterly averages of out-strength normalized by contemporaneous N. Because N grows from ~30 to ~390, the number of pairwise tests rises by two orders of magnitude while the BH threshold (q=0.05) is held fixed; detection power and network density therefore change systematically. The /N normalization does not automatically guarantee that relative out-strengths of incumbents remain comparable. Without a fixed-core or density-matched robustness check, the reported Bitcoin decline and elevated post-2022 turnover may partly reflect the expanding universe and tightening multiple-testing correction rather than pure reorganization of influence.","section":null},{"comment":"Methods §3 and Results §4: linear pairwise Granger causality on 1-minute returns is treated as a faithful measure of directed “influence.” Although the PC1 filter addresses a shared market factor, residual non-stationarity after ADF filtering, possible nonlinear dependence, and the linear VAR specification itself remain unexamined. At least one additional check (e.g., nonlinear GC or a coarser sampling frequency) is needed to support the interpretive leap from statistical predictability to market influence that underpins the hierarchy narrative.","section":null}],"minor_comments":[{"comment":"Fig. 1: power-law exponents are given for only two assets; a short table or statement that the remaining coins yield comparable γ would strengthen the stylized-fact claim.","section":null},{"comment":"Eq. (4) and Table I: the variance-reduction percentage and GX→Y are reported for a single illustrative pair; a brief distribution of typical GC strengths across weeks would help the reader gauge effect sizes.","section":null},{"comment":"Fig. 5 caption: the symbol list is helpful but lengthy; a supplementary table mapping tickers to full names would improve readability.","section":null},{"comment":"Introduction and Conclusions: several self-citations to the authors’ wallet-network papers are appropriate but could be condensed to avoid redundancy with the price-based focus of the present work.","section":null},{"comment":"Typographical inconsistencies appear in author names and affiliations (e.g., “P eyy ala”, “C hakraborty”, “Insti tutes”); these should be corrected in production.","section":null}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a solid empirical extension of existing GC-network work on crypto, but the temporal-comparability issue with a growing N is load-bearing for the headline ranking claim. If the authors can supply a fixed-core or density-matched robustness analysis, the paper would be suitable for the journal; otherwise the central narrative remains under-supported. Scope fit with q-fin.TR is good."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing: this is a careful, longer-horizon extension of the 2020–21 Granger-causality crypto networks (Scagliarini et al.). They push the window to 2020–March 2025 (275 weekly 1-min networks), let the asset universe grow from ~30 to ~390, add ADF stationarity filtering, Benjamini–Hochberg FDR at q=0.05, and then rank by quarterly-averaged out-strength. The headline empirical result is new relative to that short-window literature: ETH stays #1, BTC declines, and 17 distinct coins cycle through the top-5. That ranking time series (Fig. 5) is the paper’s actual contribution.\n\nWhat they do well is the hygiene. Stationarity is checked (99 % pass ADF), multiple testing is controlled, lag selection is BIC, and they run a PC1-removal robustness check that leaves out-strengths and link weights almost unchanged (Spearman ρ ≈ 0.999 / 0.995). The sublinear in/out-strength scaling (α ≈ 0.91, with SE and a volume covariate that does not kill it) is cleanly reported and not an artifact of liquidity. Heavy-tail exponents on the returns themselves are standard but properly estimated. Self-citation of their earlier wallet papers is light and not load-bearing.\n\nThe soft spot that actually matters is temporal comparability of the ranking. As N grows by an order of magnitude, the number of pairwise tests grows quadratically, so the BH threshold tightens while new nodes keep entering. Normalizing out-strength by contemporaneous N does not fully neutralize that. The reported BTC decline and the post-2022 turnover spike could therefore mix genuine influence shifts with changing detection power and network density. The PC1 and volume checks do not address this. Linear GC on 1-min returns is also only a statistical notion of “influence,” which the authors themselves note; that is a known limitation of the method, not a hidden flaw. No code or data release is a practical annoyance for replication, nothing more.\n\nThis is useful descriptive microstructure work for anyone who builds crypto influence or systemic-risk indicators. It does not resolve a theoretical question or change pricing practice. I would send it to peer review; a referee can ask for a fixed-core-asset robustness check and uncertainty bands on the ranks. Worth reading if you work in this niche; not a must-cite for everyone.","headline":"Solid longer-horizon GC-network extension of Scagliarini et al.; ETH-dominant ranking with high turnover is the real empirical payload, but N-growth + fixed FDR makes the time series only partly comparable.","tokens_in":13267,"tokens_out":603,"would_cite":false,"duration_ms":6842,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Weekly Granger-causality networks of 1-minute crypto returns show Ethereum as the persistent top influencer while Bitcoin declines and the top-five ranks turn over repeatedly.","keywords":["cryptocurrencies","log-return","Granger causality","weighted networks","high-frequency data","nodal out-strength","market influence hierarchy"],"falsifier":"Recompute the same weekly networks with a non-linear or transfer-entropy causality measure (or with higher-order residualization beyond the first principal component) and check whether Ethereum’s continuous top ranking and the observed top-five turnover of seventeen assets disappear.","tokens_in":13266,"feed_emoji":"📈","tokens_out":907,"duration_ms":8507,"temperature":0.7,"pith_summary":"This paper maps how influence flows among cryptocurrencies by building directed, weighted networks from statistically significant Granger causal links between their one-minute log-returns, week by week from 2020 to early 2025. The authors show that the resulting networks are highly heterogeneous: a few assets account for most of the outgoing and incoming influence, and the ranking of assets by out-strength changes substantially over time. Ethereum stays at the top of that ranking throughout the five-year window, Bitcoin’s relative position erodes, and seventeen different coins appear in the top five at least once. A sympathetic reader cares because the result replaces the image of a stable, Bitcoin-centered hierarchy with a picture of a competitive, non-stationary market in which leadership continually shifts. The same analysis also recovers the familiar heavy-tailed return distributions and documents a sub-linear scaling between in-strength and out-strength that is not explained by trading volume.","feed_headline":"Ethereum stays top influencer as crypto rankings keep flipping","feed_subtitle":"Weekly Granger networks of 1-minute returns show Bitcoin fading and 17 coins cycling through the top five","key_machinery":"Time-dependent directed weighted networks whose edges are the log-variance-ratio Granger-causality strengths (after ADF stationarity screening and Benjamini–Hochberg FDR control at q = 0.05) computed on one-minute log-returns inside non-overlapping weekly windows; nodal out-strength is then used as the ranking measure of influence.","core_discovery":"Ranking cryptocurrencies by nodal out-strength in weekly Granger-causality networks of one-minute log-returns reveals a dynamically evolving hierarchy: Ethereum remains the most influential asset across 2020–2025, Bitcoin’s relative influence declines, and seventeen distinct assets occupy the top-five positions, demonstrating a competitive and non-stable organization of market influence.","pith_inferences":["If out-strength rankings continue to turn over this rapidly, portfolio-risk models that treat a fixed set of “systemic” cryptos as permanent hubs will systematically mis-estimate contagion paths.","The same weekly Granger pipeline could be applied to traditional equity or FX markets to test whether the absence of super-stable nodes is crypto-specific or a general high-frequency phenomenon.","Combining the price-based influence networks with wallet-level transaction networks (as the authors themselves suggest) would allow a direct test of whether price leadership coincides with on-chain fund-flow leadership."],"forward_implications":["Market-influence rankings cannot be treated as stable; any monitoring system must be recomputed on short horizons.","Ethereum’s sustained out-strength dominance supplies a quantitative counterpart to narratives of its technological maturation.","Bitcoin’s gradual loss of relative out-strength indicates that market leadership is not locked to the oldest or largest asset.","The sub-linear in-strength–out-strength scaling implies an intrinsic asymmetry in how influence is received versus transmitted across the crypto market.","Seventeen distinct top-five occupants over five years quantify the competitive turnover of the ecosystem."],"fun_headline_variants":["Ethereum holds top out-strength as Bitcoin fades in weekly crypto networks","Seventeen coins cycle top five while Ethereum leads influence hierarchy","Crypto rankings flip often as Ethereum stays most influential asset","Granger networks show Ethereum dominant amid unstable influence order","Bitcoin influence declines as Ethereum anchors shifting crypto hierarchy"],"cache_read_input_tokens":128,"weakest_assumption_plain":"That statistically significant linear Granger causality on one-minute returns, even after stationarity filtering and false-discovery control, is a faithful measure of directed influence rather than an artifact of residual common factors or the linear VAR specification itself.","fun_headline_variants_meta":{"raw":{"variants":["Ethereum holds top out-strength as Bitcoin fades in weekly crypto networks","Seventeen coins cycle top five while Ethereum leads influence hierarchy","Crypto rankings flip often as Ethereum stays most influential asset","Granger networks show Ethereum dominant amid unstable influence order","Bitcoin influence declines as Ethereum anchors shifting crypto hierarchy"]},"model":"grok-4.5","effort":"low","cost_usd":0.00491,"raw_usage":{"total_tokens":1331,"prompt_tokens":721,"num_sources_used":0,"completion_tokens":80,"cost_in_usd_ticks":49100000,"prompt_tokens_details":{"text_tokens":721,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":530,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":721,"tokens_out":80,"duration_ms":5610,"temperature":1.0,"reasoning_tokens":530,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-14T17:20:41.824755+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Recompute the same weekly networks with a non-linear or transfer-entropy causality measure (or with higher-order residualization beyond the first principal component) and check whether Ethereum’s continuous top ranking and the observed top-five turnover of seventeen assets disappear.","supporting_citations":[],"review_version":2}