KarapticResearch / Index
A public research log testing published crypto trading strategies against real Binance data. Literature review → replication → gross → net-of-costs → verdict, one note per hypothesis. Closed and negative results are published alongside positive ones.
Open a note for the full, indexable abstract and links to methodology, code, data, and references.
Introduces the series and its method. An audit of roughly 200 pre-built trading strategies on a marketplace (a "CMC Skill Hub") found that most rest on the same public metrics anyone can compute for free.
Read note ↗Asks whether grid trading produces an independent trading edge, or whether its returns come from market exposure, execution economics, and market-making infrastructure. Chen, Chen & Jang derive analytically that, under a symmetric random walk, a classical grid's expected profit is zero before costs.
Read note ↗Tests whether a cross-sectional momentum spread survives on Binance USDⓈ-M perpetuals net of realistic costs, not just in an academic backtest. Liu, Tsyvinski & Wu document momentum as statistically significant across a broad crypto sample; Starkiller Capital's practitioner backtest beat the market in relative terms through the 2021–2022 bear market without an absolute profit, and named a 125-basis-point cost threshold that kills the effect.
Read note ↗Asks whether time-series momentum (well established on traditional assets by Moskowitz, Ooi & Pedersen) still holds in crypto today, or whether early favorable evidence (Liu & Tsyvinski, pre-2018 data) described a regime that has since passed. Grobys, Kolari, Sandretto, Shahzad & Äijö's 2025 study, the longest and most recent available (2016–2023, 30 large-cap coins), splits the sample by regime: strong pre-July-2020, but negative and not statistically significant in the post-July-2020 period that matters for a trader today.
Read note ↗Perpetual funding carry arrives with strong academic backing: He, Manela, Ross & von Wachter report a Sharpe of 1. 80 for a spot-short-perpetual carry trade, and Schmeling, Schrimpf & Todorov document crypto carry averaging roughly 7%/year on a fixed-date basis.
Read note ↗The literature contains two unrelated strategies both called "reversal." Branch A (Dobrynskaya): buy coins that just crashed, hold 10–12 weeks, a bubble-and-burst pattern found on a roughly 2,000-coin CoinMarketCap sample, 2014–2020.
Read note ↗Unlike the rest of the series, this note goes system-first: Klines, an event detector built from months of manual replay of Binance volume spikes, is checked against the academic literature only after it had a stable shape: eleven papers covering why volume anomalies predict returns, how exits should be managed (triple-barrier, maximum adverse excursion), and what remains genuinely open. On the larger, pre-patch walk-forward sample, the core "ALIVE" tier shows an 85% win rate and the narrower "ROCKET" tier 100%, stable across eleven weekly cohorts; a newer, more honestly timed log (real detection timestamps instead of a hardcoded offset) shows a materially weaker 63.
Read note ↗Full repository, methodology and references: github.com/Silent47boryara/searching-for-edge ↗